Improved Radar Data Clustering using Camera Data for Extended Target Tracking

This work proposes an approach for improving the clustering of measurements obtained from high-resolution radars by considering additional camera input. In a number of practical applications, such as tracking multiple closely-separated targets of varying shapes and sizes, classical density-based clustering algorithms produce erroneous results caused by merging and splitting. As a result, either the measurements originating from distinctly different targets are clustered together, or measurements from the same target are grouped into multiple clusters. Wrong clusters will lead to significant performance degradation in the tracking and classification results. In this work, a camera data-assisted improved clustering algorithm is presented. Performance evaluation of the proposed approach is performed using a publicly available dataset. Results indicate that using the additional camera data improves the clustering performance.