Abstract

With the increasing population, the issue of pedestrian safety is currently of major concern in
most cities of the world. Pedestrian safety is concerned with ensuring the well-being of pedestrians
and reducing the potential risk areas as well as implementing measures to reduce accidents. The
aim of this study is to propose a computer vision and cloud-based solution that enhances pedestrian
safety by collecting, visualizing and analyzing pedestrian and vehicular data across different
intersections in the city of Montreal. In the past, the rate of accidents in the City of Montreal
involving pedestrians has been quite high, therefore a method to solve this problem has led to this
study.
About 200,000 images were collected across 43 intersections in the city of Montreal from the
Traffic cameras – Ville de Montreal website. The data was collected from March 8, 2020, up until
March 22, 2020 and then from May 1st, 2020 to 11th May 2020. An object detection and
classification model using Faster RCNN algorithm to identify pedestrian and vehicles at the
intersection was implemented. Further, this model was used to obtain a dataset showing the
number of pedestrians and vehicles at the intersections. The information obtained from this data
set was used for visualization and in-depth analysis of the pedestrian and vehicle data in order to
derive patterns of peak and non-peak hours and high-risk intersections.
IV
Furthermore, zero inflation poisson distribution model was implemented on our dataset to
display the timings and intersections which had zero pedestrian counts for long hours of the day
as compared to the vehicle count. A heat map was generated to visualize the dataset and to assist
data viewers to identify which areas should get most attention.
Finally, we created a prototype solution that mimicked the traffic control system by utilizing
LEDs and microcontrollers (IoT device), cloud services, publish/subscribe model, and object
detection. To implement this prototype, the data obtained through the object detection model was
sent onto the cloud (Cloud MQTT), from where it was used to control the programmed
microcontrollers (IoT devices) present at the different intersections based on the vehicle and
pedestrian counts. The system managed to show excellent accuracy for detection of vehicles and
pedestrians on the dataset, and the delay experienced in controlling the microcontroller was also
negligible, thus making our system effective and reliable.