A Novel Method for Improving Point Cloud Accuracy in Automotive Radar Object Recognition

High-quality environmental perceptions are crucial for self-driving cars. Integrating multiple sensors is the predominant research direction enhancing accuracy and resilience of autonomous driving systems. Millimeter-wave radar has recently gained attention from academic community owing to its unique physical properties that complement other sensing modalities, such as vision. Unlike cameras LIDAR, millimeter-wave not affected by light or weather conditions, a high penetration capability, can operate day night, making it an ideal sensor object tracking identification. However, longer wavelengths signals present challenges, including sparse point clouds susceptibility multipath effects, which limit their accuracies. To enhance recognition capability radar, we propose GAN-based cloud enhancement method converts into RF images with richer semantic information, ultimately improving tasks detection segmentation. We evaluated our on CARRADA nuScenes datasets, experimental results demonstrate approach improves classification 14.01% segmentation 4.88% compared current state-of-the-art methods.