Research And Implementation Of Anti-Intrusion Early Warning System Based On The Fusion Of Millimeter-Wave Radar And Visual Sensor

On highways with continuous traffic and pedestrians,it is also difficult to ensure the safety of the lives of road workers and traffic policemen handling accidents.The endless number of cases is alarming us all the time,and we all want to use a practical and effective way to reduce the incidence of these accidents.The environmental awareness system for auto-driving smart vehicles can effectively alleviate the destructive effects of traffic accidents.If it is used to close the first and the end of a road segment,it can perceive the vehicles that may be dangerous in advance,and warn road workers to avoid in time within a safe time range,which can greatly reduce the chance of accidents for road maintenance personnel.Therefore,this paper presents a closed road anti-intrusion early warning system based on the fusion of millimeter-wave radar and visual sensor,By using two sensors,millimeter wave radar and visual sensors,the speed and distance information of oncoming vehicles are obtained,and data fusion is carried out in a timely manner to predict whether oncoming vehicles will pose a danger.After detecting threatened vehicles,road operators will be reminded to avoid them in a timely manner.The main work includes:(1)Considering that single millimeter wave radar target detection cannot determine the category of the target and can only obtain the relative velocity and position information of the target,it is necessary to introduce visual sensors to compensate for the shortcomings of millimeter wave radar.With the rapid development of in-depth learning,YOLO series of algorithms are very fast in detection,and have a high real-time performance when applied to road environment perception.They are most suitable for the research scenarios in this paper.Therefore,an improved YOLOv5s algorithm is proposed,which can greatly improve the detection accuracy without reducing the detection speed.(2)The accuracy of millimeter wave radar and visual sensor fusion calibration has been a long-standing problem for researchers.The prerequisite for improving the fusion calibration accuracy is that the camera calibration accuracy and millimeter wave radar calibration accuracy must be sufficiently high,so that the fusion of the two can achieve a"1+1>2"effect.Therefore,improving the calibration accuracy of each sensor itself is a guarantee for improving the fusion calibration accuracy.This article proposes a calibration method for 80mm telephoto cameras used in experimental scenarios based on the HarrisRANSAC algorithm.This method greatly improves the accuracy of corner detection when using fuzzy checkerboard images for Zhang’s plane calibration.It not only reduces the difficulty of calibration,but also greatly reduces the time consumption of the calibration process.(3)By studying the construction of environment model for multi-sensor fusion of visual and radar information,the reliability of information from different sensors is evaluated and the confidence evaluation is generated based on the independent perception of each sensor.On this basis,a fusion modeling algorithm based on D-S evidence theory is proposed,which interactively validates the information of each sensor,calculates the posterior probability distribution of multi-information entropy,constructs a"visual-radar"hybrid graph model,labels the semantics of scene image sequence objects,describes the high-order asymmetric relationship between multi-sensor variables in space-time,and solves the uncertainty problem in multi-source information fusion of visual and millimeter-wave radar.