Design and Evaluation of a Real-time Pedestrian Detection System for Autonomous Vehicles

Design and development of autonomous vehicles capable of moving safely on roads by sensing the environment has motivated researchers to focus on design of pedestrian detection systems. Similarly, Convolution Neural Networks (CNN) is considered as one of the preferred image classification algorithms. Most of the papers reported in literature employ standard object detector modules available online for pedestrian detection. Design of a real-time pedestrian detection system using CNN for autonomous vehicles is proposed and the system is designed from scratch without using any standard module available. The performance evaluation of proposed system is carried out using INRIA dataset, PETA-CUHK dataset and realtime video input. The CNN parameters were also tuned to achieve best possible recognition accuracy. Recognition accuracies ranging between 96.73 - 100% is obtained on using the system, based on dataset employed. The results obtained are also compared with the results reported in literature and the system designed can be considered on par with those reported in literature. The proposed real-time pedestrian detection system can also be employed as driver assistance system for non-autonomous vehicles.