Social inclusion of elderly pedestrians in urban contexts by enhancing healthy mobility means increase the level of walkability, where the perception of safe walking and road crossing is a crucial factor. The measurement of stress during walking and dynamic collision avoidance through the acquisition and the analysis of physiological signals during experimental activity requires the design of proper experimental settings and protocols, in order to assess innovative approaches towards an affective walkability, and open novel investigation elds involving a ective computing, pedestrian dynamics and arti cial intelligence. The main aim of the paper is to illustrate some preliminary studies conducted within an experimental activity to test the validity of the approach.
Such approach suggests new perspectives in the design/planning of the future
of the cities: how accessible, comfortable, safe and secure the city is for
walking/crossing, also in presence of perceptive and motor limitations [
Amid the main factors characterizing walkability (accessibility, comfort, and
safety), the concept of safe walking is crucial: \walking is a basic and common
mode of transport with bene ts to health and the environment. Measures must
be taken to improve the safety of walkers" [
The vehicle-pedestrian interaction while road crossing is a crucial point in the perception of safe walking, and subjective perceptions must be considered, due to the di erent degrees of vulnerability (age, gender, disabilities) of pedestrians.
In order to focus on the pedestrians' perception of safe road crossing, new
approaches could be investigated, and new sets of technological devices adopted.
Wearable sensors and the a ecting computing approach [
Sensor technology has improved signi cantly and allows measuring
physiological signals as well as registering daily life activities (e.g. inertial sensors, video
and audio recordings). Several sensors can be easily integrated into smartphones
or wearable devices [
Physiological signals are nowadays widely used to detect a ective states [
Arousal is a physiological and psychological state that can be related to sensory alertness, mobility, and readiness to respond. It is thus activated in the interaction between pedestrian and the environment as a defensive reaction to preserve safety, which is the connotation of stress here adopted.
Moreover, motion data both physiological, measuring the muscle activity with Electromyogram (EMG), and inertial (accelerometer and gyroscope data) have been adopted, in order to design a novel integrated approach in the study of pedestrian dynamics, within a multi-modal framework.
Relying on di erent signal sources that register both physiological and dynamic walking responses will provide accurate results for a ective state recognition tasks. Within this preliminary research, the multi-modal approach has been adopted, in order to synchronize and properly segment all the raw data streams collected. Depending on the type of signal, a proper noise reduction ltering has been performed. Characteristic patterns of these signals, related to induced stressful states, can be identi ed and provide new insights for measuring the perception of safe walking.
In this paper, preliminary studies conducted within an experimental activity are described to test the validity of the approach. Section 2 illustrates our study of the perception of safe walking, explaining in details in-vitro (through the design of the experimental setting in a protected space) and in-vivo (data collections in the real world) experiments and observations. Section 2.1 reports our experimental activity on human subjects during collision avoidance tasks, in a strictly controlled indoor environment, while Section 2.2 is related to the evaluation of the perception of safe side walking in a selected urban scenario and experiments considering road crossing tasks (both outdoor). Preliminary analyses of collected data are illustrated in Section 3. Finally, we draw some conclusive remarks about the next steps of the research towards the adoption of the best AI-based learning techniques to be selected to reach an a ective walkability approach through massive collection of data (from commercial sensors) supporting the design of future urban inclusive environments. 2
Studying age-driven walking and road crossing behaviors by means of direct observations, two types of experiments have been designed and developed in di erent environment con gurations, conducted respectively in Tokyo (Japan) and in Milan (Italy). These experiments involved two populations: a population of young adults (18 - 35 years old) and a population of elderly people (over 60), in order to compare di erent a ective behaviors.
In Tokyo in-vitro data collection experiments (i.e. through the design of the
experimental setting in a protected space) have been conducted in a controlled
laboratory environment, with the aim of studying pedestrian behaviour avoiding
collisions. In Milano in-vivo data collection (i.e. data collections in the real world)
in an outdoor uncontrolled environment have been carried out, to evaluate the
safety perception in pedestrian environments considering in particular crossings,
and sidewalks [
Both physiological signals and inertial data have been collected using
wearable sensors produced by Shimmer3 [
The controlled experimental environment is depicted in Figure 2. The plan of the indoor environment is reported in the top left image, showing the U path where the subjects are moving. Two subjects (sbj1 and sbj2), start at the same time. The collision avoidance zone is identi ed by a red rectangle and depicted also in the image at the top right. The two obstacles are controlled by one of the experimenter, and the two subjects have to avoid the collision (bottom right). During the rest of the U path, subjects walk with their own natural pace (bottom left). 1. Baseline acquisition: all data are acquired for one minute, the two subjects standing still. These data serve as reference for each subject. 2. Collision avoidance: the two subjects walk with their own pace, till they reach the collision avoidance zone, where they have to avoid the collisions with both obstacles and the other subject. Then they complete the U path, with their natural pace. 3. Normal walk: the two subjects walk on the second half of the U path, with their natural pace.
2.2
Perception of safe walking and road crossing
For the in-vivo data collection, an experiment in an uncontrolled urban environment has been conducted. For this experiment 14 young adults, all computer science students at the University of Milano-Bicocca, have been recruited, (20-26 years old, 7 women). The aim of this experiment is to evaluate the safe walking perception in urban scenarios, in particular of pedestrians crossings a road, and walking on the sidewalks. To this end, a two way road, in correspondence to a crossroad, without tra c lights, has been considered. In Figure 3 on the left, this experimental environment is depicted. The zebra crossing and the portion of the sidewalks where the experiment was conducted, are highlighted with red rectangles. Images on the right report a subject performing the normal walking and the crossing tasks respectively.
The experimental procedure is described as follows: 1. Normal walk: the subject walks on the sidewalks, back and forth, with its own pace. 2. Baseline acquisition: all data are acquired for one minute, the two subjects standing still. These data serve as reference for each subject. 3. Crossing: the subject has to cross the road in correspondence to the zebra crossing, back and forth. 4. Baseline acquisition: all data are acquired for one minute, the two subjects standing still. These data serve as a further reference for each subject.
During both the in-vitro and in-vivo experiments, raw data streams from
multiple sensors have been acquired. A multi-modal system has been employed to
manage all the sensors in order to synchronize and properly segment all the
raw data streams into di erent tasks. As a pre-processing step, preliminary to
the segmentation phase, each signal has been ltered to perform noise
reduction (see the review paper [
As an example of what physiological signals can reveal, the in-vivo experiment is considered.
In Figure 4, GSR and EMG signals of one subject, collected during the experiment are depicted. Event windows are also drown to highlight the three di erent tasks of the experiment: B=Baseline acquisition, W=Walk on the sidewalks, C=crossing the street. The Q window (Q=Questionnaires) is the initial part of the experiment when the subject received the informed consent, and all the details about the experimental procedure, and he/she provides personal information (age, level of instructions, gender, etc). The bottom signal (red one) corresponds to the EMG, i.e. muscle activity. As expected, muscles are activated by walking. The stops during the back and forth crossing are also clearly visible. Top signal (blue) is the EDA response. During non-stressful tasks (B and W) the physiological response denotes a low activation, (slowly varying signal), while in correspondence to the crossing tasks (C windows), signi cant peaks are detectable. Moreover, the two single crossings corresponding to the back and forth paths can be identi ed by the two main peaks. This preliminary analysis con rms that physiological signals can be adopted to detect an increase of arousal related to induced stress or activated by an attentive state originated by the interaction between subject and environment, especially during collision avoidance tasks. 4
Within this work, the problem of evaluating the perception of safe walking and road crossing (collision avoidance task) has been faced, introducing a novel investigation eld, involving a ective computing.
An a ective walkability approach that relies on physiological signals acquired by means of wearable sensors has been introduced.
The preliminary analyses on data collected in in-vivo and in-vitro experiments show that there are promising correlations between signal patterns and a ective states, related to di erent walking conditions. These considerations justify a massive e ort to collect more experimental data from multi-modal sensors within the context of the interaction between pedestrians and environment, especially considering dynamic collision avoidance tasks.
Proper features from all multi-modal signals collected can be extracted. Besides initial statistical inferences that could support our preliminary considerations, to take advantage from multi-modal sources, information fusion is mandatory, either at the feature extraction or at the decision level. Increasing collected data and introducing multi-modal fusion strategies open the direction towards the involvement of AI-based learning perspectives to reach an a ective walkability approach in the design of future urban inclusive environments, through the future massive collection of data using commercial sensors.
This preliminary research has been possible thanks to the kind support of Katsuhiro Nishinari, Daichi Yanagisawa, Kenichiro Shimura (Research Center for Advanced Science and Technology, The University of Tokyo) who supported the organization of the experiments in Tokyo, and Arch. Matteo Bel ore, supporting the team during the design of the experimental set in Tokyo.