Target Classification Using Frontal Images Measured by 77 GHz FMCW Radar through DCNN

Abstract

This paper proposes a target classification method using radar frontal imaging measured by millimeter-wave multiple-input multiple-output (MW-MIMO) radar through deep convolutional neural networks. Autonomous vehicles must classify targets in front of the vehicle to attain better situational awareness. We use 2D sparse array radar to capture the frontal images of objects on the road, such as sedans, vans, trucks, humans, poles, and trees. The frontal image includes information regarding not only the shape of a target but also the reflection characteristics of each part of the target. The measured frontal images are classified by deep convolutional neural networks, and the classification rate yielded 87.1% for six classes and 92.6% for three classes.