Towards a Real-Time Emergency Response Model for Connected and Autonomous Vehicles Yen-Hung Liu1 , Otavio de P. Albuquerque2,∗ , Patrick C. K. Hung1 , Hossam A. Gabbar1 , Marcelo Fantinato2 and Farkhund Iqbal3 1 Faculty of Business and IT, Ontario Tech University, Oshawa, Canada 2 School of Arts, Sciences and Humanities, University of São Paulo, São Paulo, Brazil 3 College of Technological Innovation, Zayed University, Abu Dhabi, United Arab Emirates Abstract Recently technological advancements in the automobile and transportation sector have gained significant interest from governments, industry leaders, and citizens. Together with Autonomous Vehicles (AV) and Connected Vehicles (CV), Connected-Autonomous Vehicles (CAV) have made a revolution in these sectors. Emergency Vehicles (EVs), such as ambu- lances, fire trucks, and patrol cars, are essential to our daily traffic life. Each of EVs has a different purpose, but all have their urgency and importance, and any time passing may cause the death of life. Thus, whenever other vehicle drivers encounter an EV on the road, they must yield to the EVs. Therefore, a CAV system that can detect EVs will significantly improve these issues. According to the Society of Automotive Engineers International (SAE), in today’s autonomous vehicles, most of them are less than Level 5, and car manufacturers assume the driver will take back control. Still, most autonomous vehicles mainly rely on their vision sensor instead of their sound sensor. Thus, when the system notifies the driver that the EVs are already close to them, it may be dangerous for the driver, pedestrians, and passengers in the vehicle. This paper proposes a conceptual framework and discusses a related methodology to support such a real-time emergency response model for CAV. Keywords connected-autonomous vehicle, emergency vehicle, emergency response, real-time response, driving assistance technology, the Doppler Effect, exceptional handling, control strategy, machine learning 1. Introduction entirely human-operated vehicles, while Level 5 vehicles are fully automated. For example, in Level 1 of automa- Recently technological advancements in the automobile tion, the vehicle may assist the driver with tasks like and transportation sector have gained significant interest steering or acceleration. Shared Autonomous Vehicle from governments, industry leaders, and citizens. Au- (SAV) should be considered at least Level 2 of vehicle tonomous Vehicles (AV) technology enables vehicles to be automation. It enables the driver to remain fully engaged controlled by precise, fast responding computers instead with the driving task but gradually transfer control from of error-prone and slowly responding human beings; human to machine. Level 2 automation features include Connected Vehicles (CV) allow infrastructure units and adaptive cruise control and automatic emergency brak- vehicles to share high-resolution information through ing. In the industry, most CAV classifies as Level 4 of wireless connectivity that can communicate to support automation, while automotive companies have carefully interaction with their internal and external environments explored Level 3. Under some circumstances, the ma- in real-time, e.g., for traffic systems and between indi- chine and the human might be sharing of controlling the vidual vehicles. Connected-autonomous Vehicles (CAV), driving task. Such cases can be dangerous for the driver which integrates the best of both AVs and CVs, have and passengers due to the spare time between control- revolutionized these sectors [1, 2]. ling exchange and human-decision making. Level 4 of According to the report of the Society of Automotive autonomous capability means cars can self-drive in most Engineers International (SAE) [3], the level of autonomy conditions without human intervention. However, there of vehicles ranges from Level 0 to 5. Level 0 concerns are many open design challenges, including technical, ethical, and regulatory matters. A completely automated THECOG 2022: Transforms in behavioral and affective computing, vehicle (Level 5) can perform all driving functions under October 2022, Atlanta, Georgia, USA all conditions. In this situation, humans are just passen- ∗ Corresponding author. gers. Envelope-Open yenhung.liu@ontairotechu.net (Y. Liu); otavioalbuquerque@usp.br (O. d. P. Albuquerque); CAV can excel in numerous advantages for smart city patrick.Hung@ontariotechu.ca (P. C. K. Hung); citizens by offering them better and more effective trans- hossam.gaber@uoit.ca (H. A. Gabbar); m.fantinato@usp.br portation services, such as dramatically reducing car (M. Fantinato); farkhund.iqbal@zu.ac.ae (F. Iqbal) crashes and driver fatigue [4, 5]. CAV offers benefits © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). for private and public transportation. It includes vehicles CEUR Workshop Proceedings http://ceur-ws.org ISSN 1613-0073 CEUR Workshop Proceedings (CEUR-WS.org) as private and service (e.g., Uber) cars, buses for public detailed analysis and enhancement of technical aspects transport (which includes school buses), and trucks (e.g. and operation, safety, and reliable performance for the garbage collectors and agricultural trucks). CAV bene- emergency environment under pressure. The proposed fits are achieved by collecting relevant information from model of CAV will have a routine for Key Performance the CAV’s context, such as geolocation, date and time, Indicators (KPIs) based on functional operational and and other individual attributes like age, address, gender, safety requirements, resiliency measures, risk analysis, and income. Therefore, CAV can infer an individual’s and Safety Integrity Level (SIL) allocation. Verification interests, traits, beliefs, and intentions. and validation will be evaluated in the related Electronic Many of today’s automated vehicles lose track of the Control Units (ECUs) with relevant standards or codes lane position when the lane markings are absent. For ex- such as the National Electrical Code/National Fire Pro- ample, erroneous lane marking recognition contributed tection Association (NFPA) 70, Canadian Electrical Code to a fatal crash of Tesla cars in California in 2018 [6]. and the International Electrotechnical Commission (IEC) CAV can address such issues since it can be connected standard and Underwriters Laboratories (UL) standard through external interfaces, like Wi-Fi, Bluetooth, Global and other codes/standards/regulations such as the work- Positioning System (GPS), and Tire Pressure Monitor- ing document provided by International Organization for ing System (TPMS). Moreover, internally, CAV works Standardization (ISO): ISO 26262 “Road vehicles - Func- with a Controller Area Network (CAN), connecting dif- tional safety” [9], and ISO 21434 “Road Vehicles – Cyber ferent Electronic Control Units (ECU) as the engine itself. Security Engineering” [10]. On the one hand, these connections are necessary to The remainder of the paper is organized as follows: provide basic (e.g., driving) and advanced features (e.g., Section 2 reviews the related scientific works present in autonomous driving and entertainment) to involved per- the literature in this work’s area. Section 3 presents a sons, such as drivers, passengers, and pedestrians. proposed conceptual cooperative framework and method- The application of CAV in real-time response for Emer- ology for real-time emergency response for Connected- gency Vehicles (EVs) has resulted in great improvements Autonomous Vehicles. Section 4 concludes the paper, in the efficiency of the process. EVs, such as ambulances, presenting a brief discussion and the found limitations. fire trucks, and police cars, are essential to our daily traf- fic life. Each EV has a different purpose, but all have their urgency and importance for emergency response and 2. Literature Review saving a life. Thus, whenever a vehicle driver encounters Safety risks may bring implications for CAV passengers, an EV on the road, the driver must yield to the EV in a other vehicles, pedestrians, and road infrastructure to safe condition. understand human aspects and perceptions towards CAV, Usually, the EV has vision and audio devices to remind such as trust [11, 12], driving style [13], and physical drivers and pedestrians of their existence, but these de- safety of pedestrians and city infrastructure [14]. These vices may sometimes be less effective in a noisy surround- numerous safety issues related to security risks may in- ing environment [7]. In 2019, 170 people were killed in fluence the consumer’s trust in purchasing CAV solutions crashes involving emergency vehicles, most of which [11]. were non-emergency vehicle occupants in the United There are many causes of EVs accidents. In personal States [8]. To address this issue, Advanced Driver Assis- factors, the EVs drivers usually drive under high pressure tance Systems (ADAS) such as adaptive cruise control, because of the time pressure, long shift hours and code blind-spot object detection, and lane departure warn- 3 running thinking. In code 3 running, the driver can ing have been designed to improve driving safety and exceed the speed limit and does not have to follow the support CAV. However, to our best knowledge, there is traffic signs in order to save the most time [15]. In en- still not much research work in ADAS for the real-time vironmental factors, drivers usually drive in unfamiliar emergency response for EVs, especially in CAV. environments or even disaster areas, and intersections This paper aims to study a real-time emergency re- are the most frequent places for EVs to be involved in sponse model for CAV (SAE Level of autonomy 3), specif- car accidents [15]. In the physical feature, fire trucks and ically for EVs, using a hybrid approach embedding either ambulances have a larger volume and are more likely to vision and sounds sensors, aiming to detect and localize cause danger when they are driven together with other the EVs by the vision and siren detection system, which vehicles. Summing up the above factors, we can under- will be discussed and analyzed the increase of accuracy stand the threat of encountering EVs on the road. Their of distance and identification measures to ensure sustain- task nature is more dangerous than general road driving, able and safe operation during normal and emergency especially since most vehicle conflicts occur at intersec- conditions and consideration of current related codes and tions. Therefore, if the traffic signal light can respond to standards environment. the situation on the road and save time on the ambulance, The paper will survey and evaluate CAV based on a it can also reduce the risk in the intersections. According station, for example. In the context of smart technol- to the task of EVs, there is usually time urgency, so for the ogy, the interface may have access to the fridge or food surrounding vehicles, the best way to respond is to slow storage information to add a stop at the supermarket or down and stop as soon as possible so that emergency grocery store so that the human can purchase supplies. vehicles do not have to be distracted by other vehicles. These contextual GPS scenarios can offer more effective However, even the Intelligent Transport Systems (ITS), itineraries for the drivers. They can include everything which include the protocols of communications between from picking up colleagues to sharing a ride for work to CAVs and the intelligent traffic, could be compromised by syncing the driver’s agenda or adapting routes to traffic cyber-attacks becoming susceptible to safety risks [16]. information, among other individual behaviors. Advancements in the CAV industry also open oppor- In previous research, the methods for detecting sirens tunities for creating a new profile of drivers. Among the from EVs can be divided into two different approaches. most promising approaches, CAV is the first alternative The first approach is to identify whether the data con- for independent visually impaired drivers [17]. Accord- tains siren sound based on the siren’s characteristics, ing to the World Health Organization (WHO), more than such as high-frequency and low-frequency, or cyclical 1 billion people live with some visual impairment in the nature [26, 27]. However, this method does not perform world [18]. WHO shows that 36 million of them are blind, well in noisy environments, especially in urban areas. and the majority of those people are over 50 years old. In- The second approach is to extract the siren signal from deed, population aging is a worldwide phenomenon that background noise [28]. For example, Fazenda et al. em- is expected to bring economic consequences [19]. Build- ployed the least mean squared algorithm to create a noise ing CAV that the elderly population can use can help the canceller to extract the target signals [29]. Nishimura et industry to overcome the economic challenge. However, al. proposed a data embedding method for the vehicle to this end, accessibility of CAV becomes imperative for location into the siren sound [30]. This research tends the sector [17]. Besides the elderly and people with visual to extract the siren signals from background noise, in- impairments, advances in CAV also open opportunities cluding lots of parts in real traffic life, by considering to enhance children’s transportation. For example, au- the Doppler Effect. The Doppler Effect is that the sound tonomous school buses may benefit their independent frequency will vary according to different speeds, so the transportation, or, even, parents may use private CAV to siren frequency in real life may differ from the spectrum drive their kids to school. we observe. Referring to the siren datasets we collected Driving essentials scenarios include, but are not lim- from the Web, Schröder et al. showed that some audio ited to, studies on Computer Vision (CV) to enhance CAV software could help to mimic the Doppler Effect in the capabilities of (partial or complete) self-driving, Artificial datasets, such as Adobe Audition 1.0 [31]. Intelligence (AI), Cloud computing, and machine learn- Our hybrid approach aims to localize the siren by ing, among other computational domains concerned to a time delay estimation method and a sound intensity enhance essential driving functionalities [20, 21, 22, 23]. probe method in the Path Planning function. For exam- Essential functionalities cover GPS services (to allow au- ple, Fazenda et al. showed that the accuracy of the time tonomous driving) and a range of sensing technologies delay estimation method is better in the distance between suitable for driver’s and passenger identification, includ- the emergency vehicle and the driver in a long distance. ing vehicle and road infrastructure detection and parking But if it is in a short distance, the sound intensity probe assistance. This scenario involves Computer Vision tech- method can also get a higher accuracy [29]. niques and route generation, which come from Computer Different projects have made great efforts to advance Science and Automation Engineering backgrounds. An- in the CAV areas. Big technology and automotive com- other technology, such as the blockchain, has already panies and universities have been working together to been embedded in CAV, considering its efficient perfor- advance the projects, designing vehicles and developing mance mechanisms for decentralized distributed storage different algorithms and drive systems split into different and security management of big data [24, 25]. levels of autonomy. The main stakeholders in this real-time emergency Uber, in partnership with Carnegie Mellon Univer- response scenario are the automobile sector, the mobile sity; Lift together with General Motors; and Didi, with application market, and average drivers and passengers. Japanese automotive companies, have been building self- An interface for this scenario can assist drivers in se- driving cars and ride-sharing services with level 4 and lecting and monitoring their driving routes, including planning to build level 5 of autonomy. Uber, Lift and contextual stops for either safety or personal matters. Didi are companies that provide mobility services and The emergency response model component may collect have a great power of traffic data collection, which is CAV’s contextual information, like the vehicle’s fuel or the key to developing and improving their automation another engine status. Then, it uses such information to system and models [32]. AutoX, in 2018, built an ad- determine if the GPS route must add a stop at the gas vanced full-stack self-driving AI platform in partnership with Alibaba Group, Chery automotive, NVIDIA, and 3. A Conceptual Framework and other companies to build SAE Level 2 and 3 assistive- driving vehicles, including RoboTaxis and RoboTrucks. Methodology The companies’ idea for the next decade is to develop a CAVs are under various scenarios and operating condi- full (Level 5) autonomous car, beginning in 2021 with the tions of residential, industrial facilities, transportation first Fully Driverless RoboTaxi Service to The Public in electrification, and grid-connected integration. There- China [32, 33]. fore, it should provide a comprehensive evaluation of the Google Waymo, in partners with automotive com- safety risks during the operation of CAV by identifying panies such as Fiat-Chrysler, Audi, Toyota, and Jaguar, hazards and estimating risks in different operating condi- has been working on self-driving vehicles of autonomy tions and modes. Codes and standards roadmap should level 4, operating the Waymo Driver, a commercial au- be performed based on fault propagation modeling and tonomous ride-hailing service, in San Francisco, Cali- analysis, simulation and evaluation will be presented and fornia. Recently in 2020, Waymo started the operation analyzed by independent protection layers for all possi- of Waymo Via, transporting commercial goods that use ble normal and abnormal operating conditions. Besides, autonomous vans and trucks [34]. The Apollo project, evaluation and validation of technical, economic, safety, Baidu’s open-source self-driving platform, was created reliability, and availability with risk factors, life cycle to test and improve the CAVs’ motion planning and ve- costing, and environmental assessment will be discussed. hicle control algorithms to aim the driving safety and The research will consider the technical requirement riding experiences. AVs such as the Lincoln MKZ Sedan of CAV for safety and performance evaluation. It will and the Ford Transit Van were used to train their dy- consolidate the currently available codes and standards namic models by real-world road data collected from of CAV and propose new evaluation criteria for real-time Apollo autonomous vehicles driving on urban roads [35]. emergency responses, to support CAV standards. CAV The Apollo platform was its 6.0 version at the end of has two aspects: hardware and software. Hardware gov- 2020. The union of efforts of the big technology compa- erns sensors such as Vehicle-to-Vehicle (V2V), Vehicle-to- nies and automakers to conceive powerful Autonomous Grid (V2G), Vehicle-to-Infrastructure (V2I) and Vehicle- Driving Systems (ADS), and consequently build fully Self- to-Everything (V2X) technology, and actuators. Software driving cars have provided great technological advance- deals with processes of perception, planning, and control. ments aimed at reaching common goals such as enhanc- V2X technology components V2V and V2I allow the vehi- ing safety, decongesting roadways, saving time for users, cle to communicate by receiving information and talking reducing greenhouse gas emissions, and ensuring mobil- to other systems in the environment. These environ- ity for all people, including the disabled and the elderly mental communication systems can be other vehicles or [36]. smart city light-changing signals. CAV should be tested To enhance situational awareness in CAV, our pro- according to their ability to transition with city speed posed model will also incorporate the research work in restrictions during an emergency. computer vision tools such as geolocalized photos and During autonomous driving, one of the most danger- videos of the situations into the proposed model [37]. ous maneuvers is lane changing. Even with ADAS, lane Furthermore, the model is expected to be implemented change is still very complex and potentially dangerous. by a machine learning approach based on Support Vec- ADAS systems should be tested on features like Adaptive tor Machine and Neural Network with the datasets we Cruise Control (ACC), Autonomous Emergency Braking collected for this research and responded to the real-time (AEB), and Lane Keep Assistant (LKA). Planning opera- input data [38]. tions should be tested under different scenarios to test the This research will consider the operation of CAVs that vehicle’s ability to adapt to road circumstances. Accord- face some technical challenges, such as the instability in ing to recent literature, the Path Planning component of some operating conditions due to the dynamic response CAV comprises three functions: Mission Planning, Be- of emergency traffic scenarios on a real-time basis. Most havioral Planning and Motion Planning. A typical task of the CAV mainly rely on their vision sensor instead of each function is outlined as follows: (1) The Mission of their sound sensor. One of the related research areas Planner: High-level decisions such as determining pickup is the hearing impaired [39, 40]. In recent years this destination locations and road selections achieve the tar- research area has gradually begun to move into CAV [36, get mission. (2) The Behavioral Planner: Dynamic ad-hoc 41]. Therefore, we believe that detecting the approaching decisions such as lane change, intersection crossing, and EVs using both a vision sensor and a siren detection overtaking. (3) The Motion Planner: Collision avoidance, system is essential in the future. obstacle avoidance, alarm generation, etc. The proposed model will be incorporated into Path Planning. 3.1. Evs Identification Process route of EVs, we will respond. If there is no conflict, we will continue to observe but do not require a response. In Figure 1, Our proposed EVs identification process will Basically, we hope we can detect both sound and visual start by detecting the approaching vehicle based on vi- detection to give complete information to the driver. If sual and sound sensors. There are usually two ways to we can not detect the precise position, our system can detect EVs. One is to see their unique appearance, and probably predict the possible position to give the most the other is to hear their siren sound. In the siren sound appropriate response. detection, we first identify which type of emergency ve- hicle it belongs to and extract the specific siren sound from the background noise. After we extracted the siren 3.2. Algorithm sound, we used the time delay estimation method and Then we introduce our Emergency Vehicles response sound intensity probe method to localize the direction algorithm. Table 1 is a brief conceptual introduction. of the sound. From the previous research [29], the time When we detect approaching EVs, we first evaluate our delay estimation method has better performance in long speed. If our speed is higher than a certain speed, we distances than the sound intensity probe method. Thus, gradually decrease the speed to maintain safety. Then we can decide the direction of the siren sound detection. when the position of EVs has been confirmed, the system starts to perform the lane change to yield to EVs. The method of changing lanes will first detect vehicles in the vicinity, according to LIDAR, which can detect vehicles within a radius of 100 meters. Result: def yieldToEVs(): Matrix = detectSurroundVehicle(); # By Using LIDAR to detect vehicles changeLane(Matrix); # change lane according to the vehicle matrix while Emergency Vechicles is detected do if speed is above certain speed then slowDown(); # slow down to certain speed if EVs’ location is confirmed then yieldToEVs(); vehicleStop(); end end end Algorithm 1: Algorithm of Emergency Vehicles Re- sponse Based on the results scanned by LIDAR, we can form a vehicle matrix according to our lane and then plan how to yield EVs based on the vehicle matrix. In Canada, the Ministry of Transportation [42], in its traditional rules, Figure 1: EVs Identification Process. stipulate that when EVs are encountered, they pull as close as possible to the right edge of the road. However, according to the real situation, it is not always the best In visual detection, we first detect the type of emer- option to pull to the right. We should make the most gency vehicle and use the image we captured and the space according to the traffic conditions. When we yield distance we detect from Light Detection and Ranging the position to EVs, the best way is to stop and let the (LIDAR) to localize the direction and the position of the EVs drive safely. After all, any vehicle movement can EVs. Combine the above information and use the GPS to cause distractions in our EVs driver. help predict the possible path of EVs and provide driver information so that appropriate responses can be made as soon as possible. If our route is interleaved with the 3.3. Conceptual Cooperative CAV system records, will be used to maintain complete traceability in dynamic and static data and properly manage risks, and By using the sharing property of the CAV, we aim to our ADAS will be reviewed according to standard secu- create a comprehensive safety standard and system. Fig- rity requirements. From the cybersecurity perspective, ure 2 is our conceptual cooperative CAV system; all the the work will follow ISO 21434 while security will be CAVs can do collaborative sense and computation. In considered in the development and deployment process, addition to the communication between vehicle and ve- embracing the Security by Design approach, considering hicle (V2V), the communication between the vehicle and requirements since the adoption of secure wireless con- the traffic lights (V2I) can also significantly save time nection protocols to the usage of encryption in informa- for the EVs to reach the destination. Usually, the biggest tion transmission. For every information transmission, problem encountered by EVs is the traffic jam on the an appropriate incident response mechanism will be initi- road or the danger encountered when executing ‘code ated, which will include methods for determining actions 3 running’, on the way to an emergency. Our EVs re- of progress or remediation and vulnerabilities analysis, sponse ADAS can collaborate with other ADAS, such as that will consider the potential damage. Adaptive Cruise Control, Autonomous Emergency Brak- ing, and Lane Change Assistant, to give corresponding responses automatically. As for the interaction between 4. Conclusion humans and vehicles, in the process of automated re- sponse, humans need to supervise and be able to inter- The operation of CAVs faces some technical challenges, vene. For example, in Tesla’s autopilot system, people such as the instability in some operating conditions due need to put their hands on the steering wheel to ensure to the dynamic response of traffic scenarios, such as emer- safety when performing lane changes. Accidents caused gency vehicle response. Therefore, the SAE classifies the by automated failures in the response of EVs can seri- vehicle from Level 0 to 5 based on the automation capa- ously affect lives. At present, a few fatal vehicle accidents bilities. In today’s CAV, most of them are in Level 3, and occur on autonomous vehicles. Therefore, before we can car manufacturers assume the driver will take back con- fully guarantee the driving safety of autonomous vehi- trol during the emergency in real time. However, drivers cles, appropriate human supervision and intervention should stay aware of automation limitations, and the man- are necessary. ufacturers should make a warning system that can give ADAS can outperform humans in some automated op- warnings far ahead of time. Nowadays, most CAV mainly erations, thus promoting traffic safety and resulting in rely on their vision sensor instead of their sound sensor. the development of autonomous vehicles. To achieve Our hybrid approach aims to detect and localize the EVs the driving safety of autonomous cars, some standard- by the vision and siren detection system. We believe ized methodologies were created. Projects such as the that detecting the approaching EVs using both vision Waymo Driver and AutoX developed the ADAS of their and sound sensors are essential in CAV, increasing the Self-driving cars using the ISO 26262 “Road Vehicles - accuracy of distance and identification measures. The re- Functional Safety,” which presents guidelines applied to search direction covers codes and standards for all control safety-related systems that include one or more electri- and communication functions in related ECUs through cal and/or electronic systems in automobiles. Due to the operating process of CAV, which should be tested for the necessity of dynamically and automatically driving accuracy and strength to support the requirements of the awareness, with lanes and vehicle detection, as well as design, development, operation, and evaluation of the collision avoidance, the CAV systems have continuously model. The requirements can fit together with standards improved their automobile intelligence and connectivity (e.g., the working document ISO 21434 “Road Vehicles – capabilities, increasing the focus on cybersecurity. Stan- Cyber Security Engineering”) and functional safety re- dardized methods of cybersecurity, such as the ISO 21434 quirements (e.g., ISO 26262 “Road vehicles - Functional “Road Vehicles - Cyber Security Engineering,” have been safety”) within the North American regulatory structure adopted, which specifies requirements for the whole life and utility requirements. cycle of automotive products of engineering-related cy- bersecurity risk management for road vehicle electrical and electronic systems and their components and inter- Acknowledgments faces [43]. This paper is supported by Research Grant Fund R20090, Our system development will comply with ISO 26262 Zayed University, United Arab Emirates. and ISO 21434. The software development life cycle in ISO 26262, from design to implantation and validation, will be followed. In addition to software development, appropriate information management, including process Figure 2: Conceptual Cooperative CAV System. References [5] M. Johns, B. Mok, D. Sirkin, N. Gowda, C. Smith, W. Talamonti, W. Ju, Exploring shared control in [1] J. Ma, X. Li, S. Shladover, H. A. Rakha, X.-Y. Lu, automated driving, in: 2016 11th ACM/IEEE Inter- R. Jagannathan, D. J. Dailey, Freeway speed harmo- national Conference on Human-Robot Interaction nization, IEEE Transactions on Intelligent Vehicles (HRI), IEEE, 2016, pp. 91–98. 1 (2016) 78–89. [6] D. Shepardson, U.s. agency to determine cause of [2] A. Ghiasi, X. Li, J. Ma, A mixed traffic speed har- 2018 fatal tesla autopilot crash, 2020. URL: https:// monization model with connected autonomous ve- www.reuters.com/article/us-tesla-crash-california/ hicles, Transportation Research Part C: Emerging u-s-agencytodetermine-cause-of-2018-fatal-tesla-/ Technologies 104 (2019) 210–233. autopilot-crash-idUSKBN1ZD24B. [3] S. O.-R. A. V. S. Committee, et al., Taxonomy and [7] L. J. Wang Wensheng, L. Sibin, Identifying the definitions for terms related to on-road motor ve- sound of ambulance whistle by frequency feature, hicle automated driving systems, SAE Standard J International Journal of Advanced Information 3016 (2014) 1–16. Technologies 6 (2012) 39–45. [4] P. Bansal, K. M. Kockelman, Forecasting americans’ [8] National Safety Council, Emergency vehicles, 2020. long-term adoption of connected and autonomous URL: https://injuryfacts.nsc.org/motor-vehicle/ vehicle technologies, Transportation Research Part road-users/emergency-vehicles. A: Policy and Practice 95 (2017) 49–63. [9] Road vehicles - Functional safety, Technical Report and transportation engineering (English edition) 6 ISO Standard No.26262:2018, International Organi- (2019) 109–131. zation for Standardization, 2018. [22] H. Gao, B. Cheng, J. Wang, K. Li, J. Zhao, D. Li, Ob- [10] Road vehicles Cybersecurity system evaluation ject classification using cnn-based fusion of vision method, Technical Report ISO Standard No. and lidar in autonomous vehicle environment, IEEE 21434:2021, International Organization for Stan- Transactions on Industrial Informatics 14 (2018) dardization, 2021. 4224–4231. [11] H. Abraham, C. Lee, S. Brady, C. Fitzgerald, [23] M. R. T. Hossai, M. A. Shahjalal, N. F. Nuri, Design B. Mehler, B. Reimer, J. F. Coughlin, Autonomous of an iot based autonomous vehicle with the aid vehicles, trust, and driving alternatives: A survey of of computer vision, in: 2017 International Confer- consumer preferences, Massachusetts Inst. Technol, ence on Electrical, Computer and Communication AgeLab, Cambridge 1 (2016) 2018–12. Engineering (ECCE), IEEE, 2017, pp. 752–756. [12] K. Lazanyi, G. Maraczi, Dispositional trust—do we [24] G. Drakopoulos, E. Kafeza, H. Al Katheeri, Proof trust autonomous cars?, in: 2017 IEEE 15th In- systems in blockchains: A survey, in: 2019 4th ternational Symposium on Intelligent Systems and South-East Europe Design Automation, Computer Informatics (SISY), IEEE, 2017, pp. 000135–000140. Engineering, Computer Networks and Social Media [13] C. Basu, Q. Yang, D. Hungerman, M. Singhal, A. D. Conference (SEEDA-CECNSM), IEEE, 2019, pp. 1–6. Dragan, Do you want your autonomous car to drive [25] T. Jiang, H. Fang, H. Wang, Blockchain-based inter- like you?, in: Proceedings of the 2017 ACM/IEEE net of vehicles: Distributed network architecture International Conference on Human-Robot Interac- and performance analysis, IEEE Internet of Things tion, 2017, pp. 417–425. Journal 6 (2018) 4640–4649. [14] A. Millard-Ball, Pedestrians, autonomous vehicles, [26] S. Kiran, M. Supriya, Siren detection and driver and cities, Journal of planning education and re- assistance using modified minimum mean square search 38 (2018) 6–12. error method, in: 2017 International Conference [15] H. Hsiao, J. Chang, P. Simeonov, Preventing emer- On Smart Technologies For Smart Nation (Smart- gency vehicle crashes: status and challenges of TechCon), IEEE, 2017, pp. 127–131. human factors issues, Human factors 60 (2018) [27] J.-J. Liaw, W.-S. Wang, H.-C. Chu, M.-S. Huang, C.- 1048–1072. P. Lu, Recognition of the ambulance siren sound [16] B. B. Gupta, A. Gaurav, E. C. Marín, W. Alhalabi, in taiwan by the longest common subsequence, in: Novel graph-based machine learning technique to 2013 IEEE International Conference on Systems, secure smart vehicles in intelligent transportation Man, and Cybernetics, IEEE, 2013, pp. 3825–3828. systems, IEEE Transactions on Intelligent Trans- [28] L. Marchegiani, P. Newman, Listening for sirens: portation Systems (2022). Locating and classifying acoustic alarms in city [17] M. Bonani, R. Oliveira, F. Correia, A. Rodrigues, scenes, IEEE Transactions on Intelligent Trans- T. Guerreiro, A. Paiva, What my eyes can’t see, a portation Systems (2022). robot can show me: Exploring the collaboration [29] B. Fazenda, H. Atmoko, F. Gu, L. Guan, A. Ball, between blind people and robots, in: Proceedings Acoustic based safety emergency vehicle detection of the 20th International ACM SIGACCESS Con- for intelligent transport systems, in: 2009 ICCAS- ference on Computers and Accessibility, 2018, pp. SICE, IEEE, 2009, pp. 4250–4255. 15–27. [30] A. Nishimura, Encoding data by frequency modu- [18] World Health Organization, Blindness lation of a high-low siren emitted by an emergency and vision impairment, 2021. URL: https: vehicle, in: 2014 Tenth International Conference //www.who.int/news-room/fact-sheets/detail/ on Intelligent Information Hiding and Multimedia blindness-and-visualimpairment. Signal Processing, IEEE, 2014, pp. 255–259. [19] G. Carbonaro, E. Leanza, P. McCann, F. Medda, De- [31] J. Schröder, S. Goetze, V. Grützmacher, J. Anemüller, mographic decline, population aging, and modern Automatic acoustic siren detection in traffic noise financial approaches to urban policy, International by part-based models, in: 2013 IEEE International Regional Science Review 41 (2018) 210–232. Conference on Acoustics, Speech and Signal Pro- [20] A. Dominguez-Sanchez, M. Cazorla, S. Orts- cessing, IEEE, 2013, pp. 493–497. Escolano, Pedestrian movement direction recog- [32] C. Badue, R. Guidolini, R. V. Carneiro, P. Azevedo, nition using convolutional neural networks, IEEE V. B. Cardoso, A. Forechi, L. Jesus, R. Berriel, T. M. transactions on intelligent transportation systems Paixao, F. Mutz, et al., Self-driving cars: A sur- 18 (2017) 3540–3548. vey, Expert Systems with Applications 165 (2021) [21] D. Elliott, W. Keen, L. Miao, Recent advances in 113816. connected and automated vehicles, journal of traffic [33] AutoX, The autox safety factor technical re- port, 2020. URL: https://medium.com/autox/ the-autox-safety-factor-c76d80e6768f. [34] M. A. Cusumano, Self-driving vehicle technology: progress and promises, Communications of the ACM 63 (2020) 20–22. [35] J. Xu, Q. Luo, K. Xu, X. Xiao, S. Yu, J. Hu, J. Miao, J. Wang, An automated learning-based proce- dure for large-scale vehicle dynamics modeling on baidu apollo platform, in: 2019 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Sys- tems (IROS), IEEE, 2019, pp. 5049–5056. [36] F. Youssef, B. Houda, Comparative study of end-to- end deep learning methods for self-driving car, Int. J. Intell. Syst. Appl 12 (2020) 15–27. [37] L. Lopez-Fuentes, J. van de Weijer, M. González- Hidalgo, H. Skinnemoen, A. D. Bagdanov, Review on computer vision techniques in emergency situa- tions, Multimedia Tools and Applications 77 (2018) 17069–17107. [38] A. Benterki, M. Boukhnifer, V. Judalet, M. Choubeila, Prediction of surrounding ve- hicles lane change intention using machine learning, in: 2019 10th IEEE International Confer- ence on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS), volume 2, IEEE, 2019, pp. 839–843. [39] F. Beritelli, S. Casale, A. Russo, S. Serrano, An auto- matic emergency signal recognition system for the hearing impaired, in: 2006 IEEE 12th Digital Signal Processing Workshop & 4th IEEE Signal Processing Education Workshop, IEEE, 2006, pp. 179–182. [40] F. Meucci, L. Pierucci, E. Del Re, L. Lastrucci, P. De- sii, A real-time siren detector to improve safety of guide in traffic environment, in: 2008 16th Euro- pean Signal Processing Conference, IEEE, 2008, pp. 1–5. [41] Y. Ebizuka, S. Kato, M. Itami, Detecting approach of emergency vehicles using siren sound processing, in: 2019 IEEE Intelligent Transportation Systems Conference (ITSC), IEEE, 2019, pp. 4431–4436. [42] Canada Ministry of Transportation, Road safety: Emergency vehicles, 2020. URL: http://www.mto.gov.on.ca/english/safety/ emergency-vehicles.shtml. [43] H. Shan, K. He, B. Wang, X. Fang, Road vehicles cy- bersecurity system evaluation method, in: Journal of Physics: Conference Series, volume 1607, IOP Publishing, 2020, p. 012054.