Traffic Target Recognition Method Based on Millimeter-Wave Radar

Zige Kang,Shihao Zhu,Haopeng Liu,Chengxin Zhang,Yong Qi,Weibin Zhang

DOI: https://doi.org/10.23919/ccc58697.2023.10240738

2023-01-01

Abstract:Autonomous driving is crucial for object detection in complex scenarios, and perception technology processes the data collected by sensors to identify observed traffic targets. Millimeter-wave radar can directly obtain depth information, which is less affected by the environment and has a strong anti-interference ability. Its measurement accuracy and stability are better than those of vision sensors, and the cost is lower than that of lidar. Based on millimeter-wave radar, this paper establishes an ensemble learning model to identify three types of traffic targets: automobiles, non-motor vehicles and pedestrians. Firstly, features were screened by correlation analysis and random forest importance analysis. A feature score algorithm is proposed to calculate the fusion results of the two analysis methods, and the sliding window algorithm is used to enhance the data. Then, based on the basic model algorithms of deep learning, combined with Bagging and soft voting method, an ensemble learning model is constructed. Finally, the experiment is carried out on the test data set. The experimental results show that the model has good recognition accuracy. Through ensemble learning, the error detection rate caused by the instability of individual learning periods is greatly reduced, thereby improving the safety of autonomous vehicles.