Research On Vehicle Recognition Technology In Front Of Unmanned Vehicle Based On Multi-information Fusion

With the increasing number of cars,an endless stream of traffic accidents have followed,which has triggered people’s research upsurge on autonomous driving and auxiliary driving technology.Environmental awareness System is one of the key components of autonomous Driving vehicle and Advanced Driving Assistance System(ADAS),and vehicle identification technology is an important research content of environmental awareness System.Real-time and accurate vehicle identification can improve driving safety and comfort.Two common sensors in vehicle identification methods are millimeter-wave radar and camera.The advantages of radar system are high sensitivity,wide detection range and strong adaptability to the environment,while its disadvantages are that there are a lot of interference signals in the detected original data,high false detection rate,and it is unable to classify and identify detected targets.The advantage of the vision system lies in its low cost and its ability to classify and recognize the obstacles ahead.Its disadvantage is that the detection range is limited by distance and it is easily interfered by environmental factors.Therefore,based on millimeter-wave radar and machine vision,this paper developed a front vehicle identification system with multi-information fusion.The specific research focuses and work contents are as follows:(1)The millimeter-wave radar can effectively select the target of the obstacles ahead and solve the problem that there is a lot of interference in the original signal.Firstly,interference targets are divided into three categories according to their causes and characteristics,and then three different methods are adopted to screen out the characteristics of these interference signals one by one.Finally,the effectiveness of each filtering method is tested to ensure that effective targets are selected.(2)Recognition of vehicle in front of video based on machine vision is realized.Based on the combination of HOG feature and SVM algorithm,the HOG feature of positive and negative sample sets is extracted respectively for the training of SVM classifier template.The HOG feature of the video image to be measured is calculated and sent into the classifier and compared with the trained template.Finally,the identified vehicle target in front is marked with a rectangular box.(3)Ahead vehicle identification with multi-sensor information fusion is realized.At the spatial level,the information fusion mainly completes the conversion between different coordinate systems,and the effective target detected by millimeter wave radar is projected onto the pixel coordinate system to generate the ROI.At the time level,the fixed frame is used in the process of information synchronization and the image data is collected at the same time point.The experimental results show that the method based on millimeter-wave radar and machine vision information fusion in this paper further improves the accuracy of identification under different weather conditions,and at the same time greatly improves the real-time performance of the vehicle identification system.