Improving Vehicle Detection using Ghost Convolution and GhostBottleneck layers in YOLOv5s

In the view of large number of vehicles on the road, manually controlling vehicles become very difficult and automated traffic management system is a need. Vehicle detection is necessary for automatic traffic management to accurately monitor and control the flow of traffic. In this paper we propose an improved YOLOv5s model for vehicle detection. It uses GhostConv and Ghostbottleneck as part of the architecture. One of the main benefits of this model is that it reduces the number of parameters, which can lead to faster training and inference times. Adapting the GhostConv and GhostBottleneck in model, reduces the amount of calculation, which makes the model faster. For training of models we make a traffic vehicles dataset that consist of four classes: auto, bike, car and bus. The experimental results show that the new model is able to reduce the number of parameters generated by 61.6% which also reduces the computation time of model. This model is also able to perform better as compared to original YOLOV5s.