Research On Preceding Vehicle Recognition Based On Millimeter-wave Radar And Machine Vision

In recent years,the research and application of the intelligent vehicle driving system has been developed rapidly,especially the perception of outside environment and identification of preceding vehicles on the road,which has been an important research field.Accurate real-time detection of the vehicle’s state information provide a strong technical support for the development of intelligent vehicle driving system and active safety system.This paper intends to identify the state of preceding vehicles in front road area,by coupling the information of millimeter-wave radar and machine vision sensor.Firstly,by analyzing the millimeter-wave radar data,the original data was classified based on the target hierarchical identification algorithm and the front target was identified based on the Kalman filter estimation.Then,the depth learning algorithm is adopted to carry on the Vehicle Identification and Road Area Segmentation simultaneously and,the coordinate mapping relationship between the millimeter-wave radar and the machine vision is established by calibrating the sensor position and parameters in experimental vehicles.Finally,the detected target of the millimeter-wave radar is mapped to the image,coupling with the vision detection results with deep learning method,the front vehicle state and position can be determined.The main research contents and results are as follows:(1)Research on vehicle identification algorithm based on millimeter-wave radar.In order to validate the effectiveness of the front target,the raw data of the millimeterwave radar was firstly identified,then the hierarchical target recognition algorithm was used to filter the effective target data,which provide the spatial distance and state information for the subsequent sensor fusion algorithm.(2)Research on vehicle identification algorithm based on deep learning.In this section,a parallel-task based vehicle detection method is designed with deep learning method,which could conduct road extraction with image semantic segmentation and vehicle detection simultaneously.The method shows good accuracy of vehicle detection.(3)Research on the Fusion Algorithm of Millimeter Wave Radar and Machine Vision information.In this section,the sensor position and parameters are calibrated by real vehicle experiments,to establish the coordinate mapping relationship between millimeter wave radar and machine vision.The sensor data synchronization in time was ensured through the backward compatible way.The results showed that depth learning algorithm can effectively split the extraction of the road area and identify the front vehicles,then the vehicle distance was mapped to the image according to their coordinates mapping relationship.In this way,the vehicle state and position information was achieved and through the real road testing,the developed fusion algorithm exhibits good accuracy.