Spatial Change Detection Using Automotive Radar

Detection of spatial change is indispensable in applications such as autonomous driving. In this paper, we present a method to detect spatial changes from a previously known sensor map of an environment using a suite of radar sensors mounted on a vehicle. In particular, this paper proposes a technique to detect a change in the position of a semi-static pole from the last measurement. We focus on feature construction as well as a supervised learning, trained using the respective features which describe the statistical similarity between the known map (M) and current radar sensor scan (S) of the same environment. In our experiments, we assessed different classification methods and feature configurations. Here, the support vector machine (SVM) trained using a combination of six statistical similarity features outperformed its competitors in classifying the change in position of a semi-static pole from the previously known sensor map (M) with an F1 score of 0.87.