Research On Target Spatial Positioning Technology Based On The Adjustment In Video Surveillance

The intelligent video surveillance is rapidly developing with the development of society, and itis a hot research topic in computer vision, which including target detection, location and tracking.Also the target spatial positioning technology is one of the hot research topics in intelligent videosurveillance, and it not only can get the target location in space, but also can provide the basis forthe three-dimensional object tracking.This thesis studies the spatial positioning of target. In order to get the spatial location of thetarget, firstly the mapping between the three-dimensional point and the two-dimensional point ofthe target should be established, but this mapping relationship is closely associated with the cameraparameters, so this thesis conduct the research from how to calibrate camera parameters. Onceobtaining the camera parameters, the spatial positioning of target can be estimated from thetwo-dimensional point in image. This thesis studies the vanishing point estimation，using vanishingpoint to calibrate the camera and the target spatial orientation, and also make some improvementscombining adjustment technology. The main work and innovation of this paper are as follows:(1) On the estimation of vanishing point, the thesis uses the method of J-linkage clustering theline segments which have the same vanishing point, and using the robust method of LMedSremoves the outlier in each set of line segments, then this thesis proposed to use the adjustmentmethod estimating the vanishing point with the effective line segment groups. Finally, theoptimization method of maximum likelihood estimation is used to estimate the optimal vanishingpoint. Experimental results show that the vanishing point estimation method which was proposed inthis thesis has a better accuracy. In order to understand the effects of camera angles on the accuracyof the estimated vanishing points, this study estimates the error distribution of vanishing pointsunder different shooting angles with the method of error ellipse.(2) On the camera calibration, the camera’s intrinsic parameters can be estimated with four pairsof orthogonal vanishing points from four images, but this result is not accuracy and stable. So thisthesis proposed an improved camera self-calibration method with the vanishing point, using theadjustment method to correct the error of the estimated internal parameters by increase2-3images.Then this thesis combines two orthogonal vanishing points and two three-dimensional coordinatesto calculate the camera’s external parameters. Considering the camera calibration may be not accurate, this thesis introduces an optimization method of camera’s parameters based on3D point.Experimental results show that the proposed method of camera self-calibration based on thevanishing point and the optimization method of camera’s parameters is effective.(3) On the target spatial orientation, after estimating the camera’s internal and externalparameters. Then the mapping relation between the two-dimensional image point and thethree-dimensional space point is established, and the spatial coordinates of the target can beestimated. Considering there has noise and measurement error in actual environment. This thesisproposed a target spatial positioning algorithom based on the method of adjustment and improvedSRUKF. By increasing the number of camera’s observation, then using the method of adjustmentand filtering to estimate the position of target in single camera and two cameras. Because the stateequation of the target spatial location is a linear equation, so the algorithm is simplified. And usingthe SVD decomposition replaces the Choleskey in the calculation of the state variable’s square rootconvariance matrix, which improving reliability of the algorithm. The experimental results showthat the target spatial positioning algorithm based on the adjustment and improved SRUKF iseffective and accurate.Finally, the summary of this thesis is made, the future research directions of this thesis arediscussed as well.