Object Detection of Pedestrian Crossing Accident Using Deep Convolutional Neural Networks

Pedestrian crossing accident recently, people injured in traffic while crossing the crosswalk. As an accident happened other caution warning systems may not be in time to safe life. Various deep learning techniques are based on a deep convolutional neural network (D-CNN) these methods are capable to fulfill object detection applications. This paper proposes the pedestrian crossing accident dataset for the detection of pedestrian crossing accidents using a video camera of front cars or dash-camera and a few CCTV videos. It presents the performance comparison between the two state-of-the-art CNN algorithms approach Faster R-CNN and YOLOv3, in the context capability to correctly classify accident, fall-down, and out-of-frame classes. This paper demonstrates that Faster RCNN outperforms YOLOv3 with a better detection in accuracy. However, the conclusion of the paper is that YOLOv3 outperforms speed to detection and has good accuracy same time able to use in real-time detection.