This study introduces an advanced illegal parking enforcement system using deep learning-based multi-object tracking for urban traffic safety. Integration of object detection, lane recognition, and collision detection addresses concerns despite traffi...
This study introduces an advanced illegal parking enforcement system using deep learning-based multi-object tracking for urban traffic safety. Integration of object detection, lane recognition, and collision detection addresses concerns despite traffic safety advancements. Validation with MOT15, MOT17, and BDD20 datasets demonstrates superior performance in static and dynamic camera environments. Key evaluations include embedding-based multi-object tracking framework performance, PAvLite-FE network accuracy and complexity, and dataset collection for multi-class object detection network training. Results indicate enhanced tracking accuracy, particularly in rapid object movements, addressing challenges like swift camera movements and small object sizes. The research concludes with the introduction of a highly efficient illegal parking enforcement system, aiming for future optimization of application software and video processing for improved cost-effectiveness.