DeepReflecs: Deep Learning for Automotive Object Classification with Radar Reflections

Abstract

This paper presents an novel object type classification method for automotive applications which uses deep learning with radar reflections. The method provides object class information such as pedestrian, cyclist, car, or non-obstacle. The method is both powerful and efficient, by using a light-weight deep learning approach on reflection level radar data. It fills the gap between low-performant methods of handcrafted features and high-performant methods with convolutional neural networks. The proposed network exploits the specific characteristics of radar reflection data: It handles unordered lists of arbitrary length as input and it combines both extraction of local and global features. In experiments with real data the proposed network outperforms existing methods of handcrafted or learned features. An ablation study analyzes the impact of the proposed global context layer.

Histogram-based Deep Learning for Automotive Radar

“…Radar-reflection-based methods are often inspired by the PointNet architecture family [1], [2]. Ulrich et al [3] use radar reflections as input to a neural network to solve an object type classification task. Schumann et al [4] tackle semantic segmentation of radar point clouds.…”

“…Our method is most related to that of Ulrich et al [3]. It solves the same task on the same data type but offers several advantages including simpler design, better robustness and increased interpretability.…”

“…Approaches that directly process point cloud data rather than their aggregated statistics have a harder time dealing with such situations. Ulrich et al [3] would need to replace the missing feature values with some default values (e. g., zeros) in order to not discard the "incomplete" points altogether. This is particularly problematic when such manipulations happen only during evaluation, introducing deviations from the typical training data patterns.…”

Self-Supervised Velocity Estimation for Automotive Radar Object Detection Networks

“…Common approaches often use classical clustering and/or tracking methods to first create object tracks, which are subsequently classified [8], [20], [21], [22]. In contrast, object detection based approaches simultaneously locate and classify objects using deep neural networks.…”

“…detection networks are still relatively new for radar, whereas they can already be considered state-of-the-art for image processing [2], [4] and lidar [3], [5]. In recent research, radar data, mainly in the form of spectra or preprocessed point clouds, are used as input to a neural network to solve various problems such as semantic segmentation [6], [7], classification [8], [9], object detection [1], [10], [11], [12], [13] or tracking [14], [15].…”