With the advancement in technology, Intelligent Transportation Systems (ITS) aims to maximize the safety and convenience of the transportation system. It focuses on the integration of technology into the traditional transportation structure for future smart cities. With the proliferation of the road network in all nations across the world, road surface condition data has become a critical component in reducing road accidents. Road condition monitoring is an important part of transportation management and affects the safety of the commute. Different methods based on manual, automatic, and semi-automatic monitoring of road conditions have been proposed in the literature. In this paper, we present a survey of smartphone-based road condition monitoring techniques. The data is acquired using smartphones and the algorithms discussed detect the road anomalies like manholes, speed bumps, potholes, and cracks, etc. A comparative analysis has been carried out based on the benefits, drawbacks, and methods used by various techniques. Furthermore, new research directions for smartphone-based detection of road surface anomalies have been presented.
The monitoring of road surface conditions has grown increasingly crucial in recent years. Road surfaces that are well-maintained improve road user safety and comfort. As a result, it is critical to regularly monitor road conditions in order to improve the transportation system's driving safety.The density of road surface anomalies is one of the key indicators used to identify road surface conditions [1]. Statistical data obtained from collected road surface information, visual field inspections, or vehicles equipped with special devices that measure and monitor road surface conditions are commonly used by municipalities. However, these technologies are time-consuming, expensive, and frequently lack the data coverage needed to provide a comprehensive picture of road conditions in large cities. Therefore, there is a needfor a low-cost, and efficient automatic or semi-automatic road surface detection technology.With the support of the Internet of Things (IoT), Intelligent Transportation Systems (ITS) employs different communication technologies to the traditional transportation system and improves the safety of road users [2], [3]. All decisions are made based on the raw data acquired by sensors or special equipment, so the data collecting step is critical.Different technologies employing laserscanners, video cameras, vibrationbased approaches, and smartphone-based approaches are being used for data collection pertaining to road surface conditions. Smartphone-based detection has emerged as a significant supplemental technology for identifying road surface abnormalities [4]. 14 12 10 8 6 4 2 Figure 1shows the statistics of the five countries with the world's largest road networks in terms of road network, population, and smartphone users [5]–[7]. It is evident that a good percentage of the population owns a smartphone. Therefore, smartphone-based road condition detection techniques could benefit the majority of road users without any additional investment.
Russia
Brazil
India
China
U.S
Based on the technique used for data processing for road anomaly detection, the smartphone-based road condition detection techniques presented in this paper have been classified into twocategories: threshold based and machine learning based techniques. 2.1
The sensor data acquired using a smartphone is analyzed to identify the patterns or values that represent the unfavorable road conditions. Different approaches have been proposed in the literature that apply thresholds to the accelerometer Z-axis data that represents vertical acceleration. Thresholding on the value of the absolute difference of accelerometer Z-axis data to detect road anomalies has been proposed in [8]– [10].The technique presented in [11] employed thresholding along with signal and image processing techniques and yields an accuracy of 93% for road anomaly detection. To overcome the limitations of static thresholding, adaptive thresholding has been employed in [12]to detect potholeswith an accuracy measurebetween 94-99%.Another technique to apply threshold on roughness index to identify road ruts has been proposed in [13]. The accuracy of the proposed method is 94%. 2.2
The researchers have employed various supervised and unsupervised machine learning techniques to process the data acquired using smartphones and identify the unfavorable road conditions. The supervised learning technique to detect road anomalies has been employed in[14]–[16]. To improve the accuracy measure, the techniques proposed by the authors have been enhanced by neural networks or signal processing techniques. The signal processing approach of Dynamic Time Warping (DTW) has been used in [17]and the system detects road bumps and potholes with 88% accuracy. In [18], authors employed deep learning techniques for object detection using the images captured by a smartphone. The proposed technique detects potholes approximately 100m ahead enhancing the safety of road users to great extent. A pothole detection technique based onConvolutional Neural Network (CNN) has been proposed in [19]. The proposed system yields 97% accuracy and uses Google API to map the detected pothole on Google Maps.
Table-1 provides the comparative analysis of the different smartphone-based state-of-the-art road condition detection techniques based on their advantages and disadvantages.
Adaptive thresholding Potholes
[27] Gaussian Model is The system ide
The smartphone-based road condition detection techniques provide a cost-effective solution using a pervasive device i.e., a smartphone. However, there are certain challenges associated with the implementation of smartphone-based systems like the placement of smartphone in the vehicle and differentiation between different types of road conditions. Different threshold based and machine learning based techniques have been proposed in the literature for road condition detection. However, to take the advantage of computationally efficient threshold-based technique and highly accurate machine-learning based approach; a hybrid technique needs to be developed to identify road conditions keeping in view the implementation cost. The communication between vehicles to transmit road condition information before hand for safety and convenience could be another area of interest. The generation of dynamic maps with updated road-condition information can be also used by authorities to streamline maintenance works. 4. References