Methods Of Camera Calibration And Region Of Interest In Vehicle Safety System

This thesis has completed by relying on the project of “Based on Multi-sourceIntegration of Mixed Traffic Pedestrian Safety Status Recognition Technology” whichis belong to the National “863” Program funded，and the project of Jilin ProvinceScience and Technology Development Plan “Mixed Traffic Video Detection Researchand Development of Smart Devices”. The paper mainly studies Camera ParametersCalibration and the method of Segmentation of Region of Interest in video detection.Since1990s, the number of our country’s vehicles and the urban populationrapidly increase, leading to traffic accident rise sharply and traffic safety issues havebecome increasingly prominent. Additionally, over the past decade the Europe and theUnited States have invested heavily on vehicle active safety mechanism and passivesafety mechanism, which further aroused our attention to security early warningsystem. The car’s passive safety mechanism can reduce the severity of the accident,and minimize pedestrian injuries suffered in an accident, but it can not avoid acollision. While, vehicle active safety mechanism can predict accident before it occurs.And the safety early warning system alert the driver to take action in advance ormandatory brake to avoiding accidents. In view of this, the safety early warningsystem’s research is very necessary, it has very important real significance in reducingtraffic accidents, protecting pedestrian safety and reducing people’s life and propertylosses in the traffic behavior.In this paper, the camera parameters calibration method’s research is mainly tomeet the demand of safety warning range, to improve the existing camera calibrationmethod. Put forward the subsection calibration method to meet the actual needs. Basedon large amount of data, we studied the relationship between the pedestrian imaging and the distance, and built the world coordinate and image coordinate mapping model,and on this basis to determine the pedestrian area of interest segmentation method. Forthe development of security early warning system laid a theoretical and technicalfoundation.The first chapter analyzes the demand of vehicle safety early warning system inthe traffic field, and the importance and necessity to develop security early warningsystem. Which leads to establish safety early warning system facing the two difficultproblems--Camera Calibration and Region of Interest Segmentation in pedestriandetection. In view of these two problems, This paper stated their research situation, andputs forward some suggestions for camera calibration method in scope therequirements, as well as the main research content in the region of interestsegmentation method.The second chapter introduces the advantages and disadvantages of the cameraimaging model, image preprocessing and corner detection methods in detail from theresearch process. This paper makes systemic research on camera calibration methods,laid the foundation to arrange suitable for vehicle safety warning requirements of awide range of pedestrian detection. In order to meet the experimental calibration andsafety early warning process, the system noise immunity requirements, through theanalysis and comparison we selected median filtering and histogram equalizationmethod for image preprocessing. And a variety of corner detection methods arecompared. we determine to select the Harris operator for corner detection, so as tofurther enhance the accuracy of experimental results, and provide the technical supportfor video detection.According to the security warning system in the range of needs, the third chapterdetermines to choose the direct linear calibration method to calibration parameters.This can meet the calibration precision, the calculation process is relatively simple.Combined with Tsai chessboard calibration method thought essence, subsectioncalibration method is proposed. Based on large amount of data, though the error evaluation we determine the precision of reliability, and ultimately determine the bestpiecewise interval, and get parameter which can accurately describe the segmentinformation.The fourth chapter put the radar data information and camera data information ontime and space integration, thus being capable to quickly determine the pedestrian areaof interest. In this paper, based on large amount of data analysis, we get the pedestrianimaging distance and corresponding pixels high relationship, pedestrian imagingdistance and pixel wide relationship, and eventually gets pedestrian imaging versus thedistance. In order to determine the pixel high impact factor, further studied cameraheight change and objects into pixel high resolution, camera and objects into pixelshigh relationship. In the end we get the accurate pedestrian interested regionsegmentation method, which is verified by the experiment, and the effect is good.The fifth chapter summarizes the paper, reviews the research process as well asthe main contents of the paper, summarizes the research progress, the innovation andapplication. Based on the summary of the existing research results, the paper make aplan about the future research direction and work.The paper’s study enriches and develops camera parameter calibration methodand region of interest segmentation in pedestrian detection progress, so that the safetyearly warning system has been further improved, not only has the theory significance,but also has practical application value.