MMW Radar Target Classification Based on Machine Learning and Ensemble Learning

Although the video-based vehicle detection technology has the merit of low-cost, there are two obvious shortages: one is that the recognition effect is greatly affected by the weather and ambient lighting, and the other is that the amount of video data is large that may lead to the extra processing time. Compared with video-based and lidar-based technologies, millimeter-wave radar has some unique advantages in vehicle detection:(1) Sending and receiving signals are not affected by weather and illumination, and can perform all-weather measurements; (2) The cost is much lower than lidar, that is suitable for popularization; (3) The data processing and power consumption are superior to the vehicle detection system based on video processing. Therefore, this paper chooses FMCW-based millimeter-wave radar to build a vehicle detection system at an intersection. Through random forest and correlation analysis, the available features are screened, and an integrated learning model obtained by integrating four kinds of traditional machine learning methods is built to classify motor vehicles pedestrians, and non-motor vehicles. The experimental results show that the model has good classification accuracy. The false detection rate caused by the instability of individual learning periods is greatly reduced through ensemble learning, thereby improving the safety of autonomous vehicles.