Research On Vehicle Detection Method Based On Information Fusion Of Millimeter Wave Radar And Machine Vision

With the advancement of technological innovation,Advanced Driving Assistance System(ADAS)is increasingly used in cars,which can effectively avoid traffic accidents.How to accurately and quickly sense the target information in front of the vehicle is the difficult point of ADAS and intelligent driving research nowadays.This paper is based on millimetre wave radar and machine vision for information fusion,to achieve accurate and stable detection of vehicles ahead,the main research content is as follows.Firstly,by analyzing the advantages and disadvantages of existing target detection algorithms,the YOLOX algorithm is established as the basic framework for vehicle detection.The CBAM-G attention module is added to YOLOX to enhance the attention to the features of the region of interest;the target context feature fusion module is proposed to enhance the features of this target with the corresponding context features of the target,which is helpful to improve the detection effect of multi-scale targets;considering the small percentage of small target samples and the small area occupied,the loss ratio of small targets is low,and the improved IOULOSS is proposed to enhance the loss ratio of small targets.The experimental results show that the improved YOLOX algorithm has better detection performance and meets the needs of complex traffic scenarios.Then,millimeter wave radar data are received and parsed,data preprocessing is performed,and the effective target cycle method is used to complete the determination of the effective radar target;the improved YOLOX model is used as the target detector,the Deep Sort multi-target tracking algorithm is used to track the target obtained from the detection model,and the appearance feature extraction model is trained on the Ve Ri776 dataset,and the correlation-matched target is used as the output target of the vision sensor.Finally,the principle of multi-sensor information is analysed and a decision-level fusion hierarchy is used for multi-sensor information fusion.The spatial fusion of sensors is completed through calibration and coordinate conversion,and temporal synchronisation is performed to determine the final output target using the Io U method and target ID matching confirmation strategy.The effectiveness of this paper’s fusion method for forward vehicle target detection is verified by building an experimental platform with millimetre wave radar and vision sensors and analysing the experimental results with single and multiple sensors.