Edge computing for traffic surveillance

Abstract

With increase in urbanization, there’s a need for better transportation systems, in terms of their efficiency. To do this, an important technique is to use the running of computer vision algorithms for identifying obstructions when they come, and notify vehicles for it in real-time, this real-time detection and subsequent alert would need computation resources located logically and physically close. In this work, we have proposed to utilize edge compute devices with lightweight GPUs which are de-coupled from the cameras. We focus our work on efficiently placing the cameras and scheduling these to the edge computing devices.