Pedestrian and Vehicle Monitoring from Traffic Videos

Pedestrian Monitoring:

This project is an attempt to develop a computationally efficient real-time pedestrian detection and flow monitoring system for mixed heterogeneous traffic scenario in the developing countries like India. Rapid urbanization and increasing road traffic have paved the way for the design of an efficient traffic management system. However, the requirements of pedestrians as road users have generally been neglected in these cities and there has been a bias of traffic control and road space allocation towards vehicular operations. Pedestrians constitute the most vulnerable group of road users and pedestrian crossing areas have been identified as one of the high-risk locations for traffic accidents. Despite various efforts, fatal pedestrian accidents still exhibit an increasing trend, 10.5% in 2016 in lieu of 9.5% in 2015 of all fatal road accidents, in urban India. We present a computer vision and queuing theory-based technique to detect pedestrians and compute key macroscopic statistics of pedestrian traffic. The proposed method has been tested on the traffic videos taken on the streets of Varanasi, India to showcase its effectiveness. The whole pipeline is computationally efficient while not compromising on performance metrics. Our proposed method can further be implemented in-site at the traffic monitoring camera sites to realize a distributed traffic management system. Such a system will be robust, scalable, autonomous and can provide real-time assistance for pedestrian control.

Figure. Results of Pedestrian Detection on the a) Region of Interest of Varanasi Dataset b) whole frame of Varanasi Dataset c)whole frame of PETS2009 Dataset d) Region of Interest of PETS2009 Dataset. The yellow boundaries indicate the polygonal ROI that we have selected.

Vehicle Monitoring:

The objective of this project is to Detect and count the number of vehicles in a given urban traffic video where the cctvs are placed at the top or on the side (with an oblique view) of the road. The motivation behind vehicle detection and counting system is ensuring road safety and facilitating medical emergencies. The workflow can be described by the following diagram:-

Figure. Results of Vehicle Detection on the Kolkata dataset at two different time instance 'a' and 'b'

Publications:

1.Ritwika Chowdhury, Kinjal Bhattacharya, Sudipta Mukhopadhyay, Bhargab Maitra, “Multi-directional Pedestrian Detection and Flow Monitoring from traffic videos”, Transportation Research Board, Washington D.C., USA, 13 Jan-17 Jan, 2019