Research On Moving Target Localization Technique Of Three-dimensional Space Based On Camera Calibration In Video Surveillance

With the increasing complexity of the monitoring environments, target tracking based on thetwo dimensional image coordinate system can no longer meet the demands for continuous andreliable tracking in intelligent video surveillance.Since each image acquired by the camera dependson its position,orientation,environmental condition and other factors,even for the same scene imagesobtained are different with different cameras.However,the most reliable feature of the moving targetis its three dimensional coordinates in multi camera target matching.Therefore, moving targetlocalization technique of three dimensional space has become one of the key technologies for targettracking in multi camera system, and the study has also been paid more and more attention.The paper is studied based on the accurate detection of moving targets, and then establishes aunified world coordinate system and computes the camera projection matrix by camera calibrationto locate the spatial location of moving targets. The paper proposes two kinds of spatial locatingmethods—the relative space position and the absolute space position.The main contributions andinnovation points are as follows:(1) Surf algorithm is researched to extract feature points, and then applies three steps matchingmethod to remove mismatching points to calculate the fundamental matrix between two cameras. i.e,first using the ratio of the feature points between the closest distance and closer distance as acriterion for primary matching. And then.counting the histogram of the difference of the maindirection of the rest matching points, and10%around the corresponding maximum bin areconsidered as matched points.Finally applying LMeds methods to make the robust matching tocalculate the fundamental matrix.(2) A new hybrid optimization algorithm combining particle swarm algorithm and geneticalgorithm is proposed to calculate the camera internal parameters derived by Kruppa equations. Andthe essential matrix is decomposed to acquire the camera external parameters.The three dimensionalspace relative coordinate is calculated using the triangular positioning principle by the projectionmatrix according to the image coordinates of the moving object.(3) In order to calculate the absolute three dimensional space coordinate.First templatecalibration method is used to calculate the main point coordinates and the aspect ratio.And then Tsaicamera model is adopted and GPS is introduced to establish the world coordinates system. Amoving target with GPS enabled mobile device moves into the FOV of a camera，and the GPS coordinates and the image coordinates are recorded respectively.Camera calibration is performedwhen more than five reference point pairs are recorded, and then calculate the projection matrix.Tomake sure the accuracy of the positioning, optimization method is proposed to obtain accuratecamera parameters.Thus, the absolute three dimensional space coordinates of moving targets in theplane can be acquired through the projection matrix and new moving target image coordinates.The thesis carries simulation experiments through serveral video sequence. Experimentalresults show that the fundamental matrix optimization algorithm can get a more accurate estimate ofthe results, and the errors of the relative spatial coordinates and absolute spatial coordinates of thetargets are in a certain range, which explains the availability of the proposed methods.Finally, the summary of this thesis is made, and futher direction on the research directions isdiscussed as well based on the work of this study.