A Research On Rear-vehicle Detection And Tracking Algorithm Based On Vision

The car is the most important human inventions of the Nineteenth century. Today,the global vehicle population is more than1billion and will continue to increase,which will lead to environmental pollution, traffic safety and other issues. Intelligenttransportation system is composed of intelligent vehicles with automatic drivingfunction, and schedules all vehicles by the control center. Control Center can give themost optimal route guidance to vehicles under the premise of ensuring globaloptimization path, thus benefits the economy and environment most.An intelligent vehicle is defined as a vehicle enhanced with perception, reasoning,and actuating devices that enable the automation of driving tasks such as safe lanefollowing, obstacle avoidance, overtaking slower traffic, assessing and avoidingdangerous situations, and determining the route. U.S. Google driverless car has beendeveloped to obtain a driver’s license, which means unmanned vehicles have officiallyentered the human life. In November2012, the "lion on the3rd" smart car developedby the Academy of Military Transportation and other units completed the firstunmanned intelligent driving test on the Beijing-Tianjin high-speed way, whichmarks that China’s smart car technology has reached a new milestone.Detection and tracking the rear-vehicle based on visual, is an important branch ofintelligent vehicle technologies. In this paper, a vision camera mounted on therearview mirror of the vehicle is used to capture the road scene information, and then,the information will be used as an input to detect, track and rang the rear-vehicles byimage processing and pattern recognition technologies.Real driving environment is always complex, buildings, pedestrians, flowers andtrees, billboards challenge the vehicle detection technology. To reduce the impact of those interferences and reduce the search region, this paper proposes a markedwatershed algorithm to select a region of interest （ROI）, only vehicles in this regionwill be detected.In this paper, a feature-based vehicle detection technology is proposed to identifythe rear-vehicles. First, the original grayscale image is sent to the binarization,exclude interference shadow line, the first line merger, the second line mergermodules to identify candidate vehicle shadow line; Then, a candidate area based onthe shadow lines will experience the symmetrically and information entropyverification module; Finally, edge detection, edge filtering, edge counting, roofposition determination and histogram projection methods will be used to determinethe area of the vehicle.Vehicle detection based on a single-frame image is not always satisfied, or evenmiss some vehicles when the traffic scenarios is complex. But the similarity betweensuccessive frames of video information is higher, so the actual position of the samevehicle in the images is stable. Kalman filtering is a method to calculate the predictedvalue of the current frame based on the estimated value of the previous frame and thecurrent frame, this paper uses Kalman filter to track vehicle.Distance between the target vehicle and the camera is important parameter forvehicle safety warning, lane change and path planning. Camera calibration techniqueis applied to get camera’s internal and external parameters in this paper, then using thecamera’s internal parameters to correct the distortion. At last, the inverse perspectiveprojection principle is applied to transform the two-dimensional image into athree-dimensional space, therefore, the exact distance is acquired.Finally, this complete system is tested in a real road environment. Experimentalresults show that the proposed algorithm can achieve good results both on city roadand beltway.