Motorcycling-Net: A Segmentation Approach for Detecting Motorcycling Near Misses

Abstract

This article presents near misses as corrective and preventive measures to safety events. The article focuses on the risk factors of commercial motorcycling near misses, which we address by proposing a near miss detection framework based on a hybrid of YOLOv4-DeepSort and VGG16-BiLSTM models. We employed YOLOv4-DeepSort model for the detection and tracking tasks, and the tracked images and identity information were stored. The sequence of image was fetched into the VGG16-BiLSTM model for extraction of image feature information and near misses recognition respectively. Video streams of near miss datasets containing motorcycling in different scenes were collected for the experiment. We evaluate the proposed methods by testing 444 sequential video frames of motorcycling near misses in urban environment. The detection models achieved 96% accuracy for motorcycle, 89% for car, and 81% for person with lower false-positive rates on the test datasets while the tracking models achieved 34.3 MOTA on the test set and MOTP of 0.77. The results of the study indicate practicality for automatic detection of motorcycling near misses in urban environment, and it could assist in providing resourceful technical reference for analyzing the risk factors of motorcycling near misses. The research contributions are: (1) A hybrid of YOLOv4 and DeepSort model to enhance object detection and tracking in a complex environment and (2) A hybrid of YOLOv4 and DeepSort model to optimize the extraction of image feature information and near misses recognition respectively for overall system performance.