Evaluation of Recovery Accuracy of Vehicles Flow Characteristics by Video Sensor Views

Traffic analysis using computer vision methods is an integral part of modern intelligent transport systems. Macromodels use a hydrodynamic approach, which requires intensity, density, and velocity versus time. Therefore, traffic counting based on a video surveillance system, is one of the urgent tasks. However, this issue still remains a challenge, especially with different video camera angles placement scenarios. In this paper, using the method of virtual detectors, we investigate how the video angle affects the count of vehicles for traffic analysis problems. Two video streams of highways of the city of Moscow in profile and front, received from open-access and own cameras, are taken as experimental data. Software to capture the intensity, density and velocity of traffic flows in real time has been developed.