An IoT approach for efficient overtake detection of vehicles using H264/AVC video data

Abstract

In this paper, a low-cost embedded autonomous system is presented. Its main device is a Raspberry Pi 2, which allows real-time analysis of traffic videos. However, the previously developed computer vision algorithms do not work at the pixel-level, meaning that they do not analyze each pixel on every frame of the video because this would entail working with such a large amount of information that it would not be possible to obtain real-time processing. Therefore, as an alternative, our system works on compressed video. In particular, our system takes as an input the H.264/AVC motion vector data produced in the encoder of the GPU embedded in Broadcom’s BCM2836 SoC. Furthermore, apart from working with very little information, the algorithms that are employed for video analysis must be efficient on their own. Thus, we propose algorithms that are capable of obtaining results directly from the analysis of a group of statistic measures, which are calculated based on a combination of input vectors. Experimentally, our system applies on the identification of overtakes, particularly those exceeding the maximum permitted speed, generating at the same time a report that is sent to a complaints e-mail account.