Deep Learning-Based Passenger Counting System Using Surveillance Cameras

Advancements in technology have led to the widespread use of the automatic passenger count (APC) system. APC facilitates informed decision-making for passengers in terms of pre-trip planning and optimal usage of transit vehicles for the operators. To make this aspect of passenger mobility efficient and convenient, methods that focus on video-based passenger counting systems are required. This paper proposes a passenger counting model using existing surveillance cameras in transit vehicles. For this, the YOLOv8 object detection algorithm is used with the deepSORT tracking algorithm. The state of the door is identified using the Hough Transformation and Canny edge detection methods. The proposed algorithm is applied to videos from the CCTV of the buses in Bengaluru. The output from the demonstration indicates that the proposed method performs better under all possible conditions.