Pedestrian Detection and Movement Direction Recognition with Convolutional Neural Network

Pedestrian detection and movement intention recognition is one of the most challenging task and active research area in Advanced Driver Assistance Systems (ADAS). It is important to detect the pedestrian and estimate future direction for safe integration of self driving cars on the existing roads. This paper proposes a system that would detect the pedestrians on the road or surroundings, and automatically recognize the direction in which they are moving viz. front, left and right by an end-to-end Convolutional Neural Network (CNN) based approach. We propose to tackle this problem as a two step approach. Firstly, pedestrians are detected by the advanced YOLOv3 architecture. The detection results are processed to recognize the movement direction. As real time performance is desirable for practical applications, we propose to use small CNN architectures like SqueezeNet, MobileNetV1, MobileNetV2 and ShuffleNet in the system. The proposed approach outperforms existing approaches in literature and provides a remarkable contribution to ensure safety in intelligent vehicles.