Faster RCNN based Vehicle Detection and Counting Framework for Undisciplined Traffic Conditions

With recent development in deep learning and computer vision techniques, Intelligent Transportation System (ITS) has emerged as a useful tool for traffic management, congestion control and building a sound traffic infrastructure. For effective monitoring of traffic conditions, there is a need to detect and count moving vehicles. Previously, several computer vision techniques have been proposed for vehicle recognition. However, majority of the techniques were not able to cover undisciplined traffic conditions. Moreover, these frameworks do not include the local vehicles of South Asian nations like Pakistan, Bangladesh and India. Keeping in view the limitations of existing frameworks, this paper presents an efficient vehicle detection and counting model for undisciplined traffic to enhance the efficiency of ITS. A dataset of more than 1200 images of vehicles has been collected comprising of 6 categories of local vehicles considering undisciplined traffic conditions to ensure robustness in vehicle detection and counting system. Faster RCNN algorithm has been used to detect moving vehicles. The experimental results show precision of 82.14% in terms of mAP.