The Research Of Optimizing Techniques For Target Space Positioning In Video Surveillance

Intelligent video surveillance is a very active research area in computer vision. Its mainpurpose is to automatically detect, track, locate, identify targets of interest, as well as understandingand analysis their activities, so as to extract useful information from large amounts of informationwhich are obtained by surveillance cameras. Determining the spatial position of the target is animportant research direction in video surveillance. By determining the spatial position of the targetto track the target and conduct behavior analysis with higher reliability, so target’s spatialpositioning attract people’s more and more attention.This thesis study the camera calibration and the method of determining the spatial position ofthe objectives, get the parameters of the camera and the conversion relationship of the coordinatesystems based on camera calibration, and determine the target position in space use the cameracalibration parameters according to a certain projection model, get accurate positioning byoptimization algorithm. About the relative spatial position of the target based on epipolar geometryconstraints adopt the bundle adjustment method, about the absolute spatial position of the targetbased on GPS adopt the Kalman filter method. The main research content and innovation are asfollows:(1) In the aspect of determine the relative spatial position of the target based on epipolargeometry constraints, use robust methods to obtain the fundamental matrix and the correspondingmatching points, use the obtained intrinsic camera parameters and fundamental matrix to computethe essential matrix, then get the accurate camera extrinsic parameters by decompose of essentialmatrix and calculate of the spatial point’s depth information, use the obtained intrinsic and extrinsiccamera parameters and the corresponding matching points of the target according to the method oftriangulation to get the target’s corresponding three-dimensional point coordinates, finally bundleadjustment method is used to simultaneous adjust the camera’s intrinsic and extrinsic parameters,also the target’s corresponding three-dimensional point set, thus determine the relative spatialposition of the target accurately.(2) In the aspect of target position determination based on GPS absolute spatial in single camera,in this thesis’s camera calibration, by set GPS coordinates as measure variable, put the cameraintrinsic and extrinsic parameters as state variable and set the state equation and measurementequation of the Kalman filter based on camera calibration model, consider the character of the measurement equation is nonlinear equation, this thesis studied camera calibration based onextended Kalman filter and unscented Kalman filter respectively, finally, according to the obtainedcamera calibration parameters to determine the absolute space position of the target in GPScoordinate system in certain precision.(3) Based on the above work, further study the absolute spatial position of the target inmulti-camera, set the spatial coordinates of the target point in GPS coordinate system as statevariable, set the target’s image coordinate in multi-camera as measure variable, establish themeasurement equation of the Kalman filter based on camera calibration models, consider thecharacter of the measurement equation is nonlinear equation, respectively use the method ofextended Kalman filter (EKF) and unscented Kalman filter (UKF) to determine the absolute spaceposition of the target in GPS coordinate system accurately.This thesis uses surveillance video sequences and real measured data to conduct simulationexperiment. Experimental results show that, in the aspect of determine the relative spatial positionof the target, the intrinsic and extrinsic camera parameters estimated by this paper’s method withhigh accuracy, can obtain more accurate spatial position of the target; In the aspect of determine theabsolute spatial position of the target use single camera, this paper’s camera calibration methodbased on extended Kalman filter (EKF) and unscented Kalman filter (UKF) can improve theaccuracy of camera calibration, and both methods have similar calibration accuracy, also verify thatthe improved accuracy of GPS receiver can improve the accuracy of camera calibration, singleperspective’s target positioning error based on Kalman camera calibration parameters within acertain error range, in the aspect of determine the absolute spatial position of the target usemulti-camera, use Kalman filter method according to the camera calibration parameters of differentpositions can improve the accuracy of target absolute positioning.Finally, the full contents of the thesis were summarized and future work is prospected.