Combined Object Detection and Tracking on High Resolution Radar Imagery for Autonomous Driving Using Deep Neural Networks and Particle Filters

This paper presents a novel approach for target detection in radar imagery, which combines an object detector and a multi target particle filter tracker. Object detection is implemented using deep neural networks, as opposed to the traditional radar object detection methods. This technique is applied to a dataset collected with a 79 GHz FMCW radar mounted on a vehicle. In this approach, object detection and tracking of roadside objects are performed in an alternating fashion to reduce the computational load required by the real time processing. The results and the thorough analysis of the parameters showed that this approach is feasible and can be successfully utilised in radar imagery for autonomous driving.