Edge pre-processing of traffic surveillance video for bandwidth and privacy optimization in smart cities

Traffic surveillance is essential for regulating traffic and reducing congestion. Nonetheless, the video data generated is non-structured, high-volume, bandwidth intensive and also raises privacy concerns. Edge computing solves these issues by preprocessing the data locally. In this paper we propose a fast, realtime vehicle detection and tracking method running on an edge device. The device sends only a single high resolution image of each detected vehicle accompanied with its trajectory. Allowing not only lower bandwidth requirements for smart city cameras, but also improved privacy, as only minimum required data is sent for post-processing. We implement object detection and tracking on an edge device, validate it in a real city environment and compare it to a state-of-the-art deep learning method using a publicly available benchmark data set where our method shows speed increase of up to 24 times.