Improved YOLOv5l for vehicle detection: an application to estimating traffic density and identifying over speeding vehicles on highway scenes

Vehicle detection and counting are getting progressively significant in highway administration due to the varying sizes of vehicles. This paper proposes a vision-based vehicle detection and counting framework. Initially, the COCO and BDD100K datasets are trained, employing improved YOLOv5l algorithm, using GhostBottleneck, for vehicle detection. Subsequently, the Centroid Tracking Algorithm counts the vehicles in videos uniquely. Later, vehicles are categorized as Light and Heavy vehicles. The model is evaluated over UCSD dataset for traffic estimation. Then, the speed of the recognized vehicle is computed to identify overspeeding. Lastly, number plate identification of the over speeding vehicles is performed. Vehicle detection performed well in precision, recall, F-score, and mAP. The centroid tracking algorithm had the best multiple object tracking accuracy and precision for vehicle tracking. The proposed model performed well in terms of detection rate, false alarm rate, and mean time to traffic detection.