Pedestrian Detection and Vulnerability Decision in Videos
Swadesh Kumar Maurya, Ayesha Choudhary · 2019 Second International Conference on Advanced Computational and Communication Paradigms (ICACCP) · 2019
In this paper, we propose a real-time framework for vulnerable road user detection from outside looking camera mounted on the vehicle. Pedestrians and cyclist are more vulnerable towards accidents on the road with the vehicle, therefore we focus on detecting the pedestrians, tracking them and predicting their vulnerability towards possible accidents with the moving vehicle observing the scene. In our framework, as the vehicle moves, the camera captures the image and pedestrians are detected using the existing deformable parts based model (DPM) with an active part selection approach. Once a pedestrian is detected then for the next n frame we use obtained spatial information in the temporal scene to track the pedestrian that ensures new pedestrians entering the scene as the vehicle moves are also get detected, the DPM is used periodically. Our system predicts the vulnerability of the pedestrians based on their path of movement with respect to the vehicle. The driver is warned if the pedestrians vulnerability of collision is high low or medium. Tracking exploits the spatiotemporal relation between frames, thereby reducing the computational cost and improving the overall runtime performance of the system. Experimental results on publicly available pedestrian dataset show that our system is capable of robustly detection, tracking and finding the vulnerability of the pedestrians to alert the driver for taking the necessary action before any collision or accident occurs.