Use of TensorFlow and OpenCV to detect vehicles

Streets are becoming overcrowded, especially in larger cities. Traffic monitoring and control is a solution to mitigate this problem, but is essential to have actuated information about vehicles. In this project the aim is to build a traffic monitoring system that detect the movement of vehicles, to observe and to count the different categories. The real-time processing (15-30 fps) of video streams works mainly in daylight. The system consists of three subsystems: image processing, motion detector, control and display. To maximize the detection speed, each subsystem can be processed on a different thread. Image processing is partially performed via OpenCV with a data set previously trained with TensorFlow. The monitoring area will be marked with a polygon. It is possible to filter sidewalks and the opposite traffic direction, so the system will avoid processing in those areas.