A Deep Learning-based Approach onVehicle Detection, Tracking, and Counting System

Transportation is one of the major features of civilization. Ever since the human race evolved, the need to move from one place to another has become mandatory. Currently, there are a lot of ways of transportation available. Yet, most of the population prefer roadways as it is simpler and more convenient. Due to the alarming increase in population, the number of vehicles on the road has also increased. Thus, it becomes tougher for the traffic police and the government to monitor the movement of all vehicles at different places. To resolve this issue, this study aims in finding the best Deep Learning (DL) algorithm that can be used to detect, track and count the number of vehicles from a video. For this process, a dataset consisting of 47 mp3 videos of the road with or without vehicles in every frame is collected from Kaggle. The dataset is then preprocessed. The preprocessing includes the conversion of video to frame-by-frame images. Two DL models will be developed using two different algorithms. The algorithms include the faster RCNN algorithm and the YOLO algorithm. The models are then trained using the images taken from the videos. The trained models are tested by analyzing various factors. The factors include precision, recall and the average IoU. The values are also compared in the form of a graph for better results. The performance of both the models is recorded and examined to find that the YOLO algorithm is a better algorithm than the faster RCNN when it comes to vehicle detection and tracking.