Traffic Conflicts Analysis in Penang Based on Improved Object Detection with Transformer Model

Current road safety detection and analysis tends to focus on hit-and-run accidents that have already occurred, while ignoring near-misses may pose a potential risk. A monitoring method for based the improved YOLOv7 of transformers is proposed in this work. First, backbone network using G3HN structure (gn Conv) with recursive gate convolution. Second, global attention mechanism (GAM) added probe improve recognition accuracy. Finally, object results are inverted by inverse perspective mapping (IPM) obtain centroid object, then probability near miss calculated analyzed DN-based, PICUD-based, PSD-based methods. This experimentwas POL37 Closed-Circuit Television (CCTV) dataset from Penang, Malaysia. The experimental show algorithm paper can effectively identify small targets phase accuracy 94.8%, smaller models, faster training speed, fulfill surveillance task near-miss events CCTV scenes.