Intelligent Measurement Of Vehicle Speed And Trail Based On Binocular Stereo Vision

With the development of social economy and information technology,the increasing pressure of road traffic makes the demand for intelligent transportation network more and more urgent.Measuring the driving state of motor vehicles is one of the most important and basic functions of ITN.At present,the commonly used methods of radar speed measurement and ground induction coil speed measurement are more or less limited in the angle of speed measurement,complicated in installation and construction,high in cost,and unable to simultaneously measure multi-target speed.By analyzing and comparing the shortcomings of traditional speed measurement technology,this paper proposes a vehicle speed measurement system based on stereo vision.A calibrated binocular stereo vision system captures two videos with certain parallax.An optimized SSD network is used to detect the license plate in the captured two videos.The detected license plate targets are tracked and stereo matched to quickly extract stereo matching point pairs.Then the calibration parameters are used to calculate the accurate three-dimensional coordinates of the stereo matching points to the corresponding space points.Finally,the vehicle speed is measured according to the distance that the vehicle passes in a certain time.Vehicle moving direction is achieved by describing the connection of selected three-dimensional points across multiple frames.The system realizes non-intrusive and non-detectable vehicle running state measurement,and overcomes the problem of multi-lane multi-vehicle speed measurement in different directions and different motion states.The system has been tested in five different practical situations,and the relevant real speed is obtained by professional satellite velocimeter.The measured velocity error ranges from[-1.6,+1.1]km/h,with a maximum error rate of 3.80%,which meets the application requirements of the national standard GB/T 21555-2007 with an error of less than 6%.The experimental results show that the method has high accuracy and validity.In addition,in order to improve the compression rate of surveillance video,this paper also proposes a CU partitioning strategy for ROI based on HEVC video coding standard,which regards people and vehicles as ROI regions,and greatly saves the compression time on the premise of guaranteeing the video quality.