Research On Target Recognition And Motion Estimation Methods Based On Color Segmentation

With the increasing of vision-based perception and measurement technology areapplied in the field of industrial, medical and public safety, target recognition andmotion estimation technology is becoming one of hot topics in the field of computervision. In this paper, we research on target recognition and motion estimationalgorithm for indoor applications. A passive cooperative target recognition algorithmbased on color segmentation is proposed for stability and accuracy recognition of thepassive cooperation objectives signs in indoor lighting conditions. A precision weightconstrained target motion estimation algorithm is proposed to estimate high-precisionmotion of passive cooperation target. A large number results of simulation andpractical experiments are shown to verify the effectiveness of the proposedalgorithms.The reconstruction equation is derived by fussing Kinect camera depth recoveryand RGB image and the feasibility is analyzed theoretically. Experimental system ofand double Kinect cameras are designed and built, and the stability experiments of thetwo experimental systems are tested. Experimental results of the monocularexperimental system show that it is difficult to obtain complete scene depthinformation stably, so the binocular experimental system is selected and built to carryout the research work with accuracy and reliability requirements.A passive cooperative target recognition algorithm based on color segmentationis proposed using priory knowledge that the indoor lighting conditions is stable andimage colors are changed small. Solid circle is utilized as signs of passive cooperativetarget with the geometry consistent in all directions and each sign is coded bydifferent color. By setting the color threshold value of the signs in HIS color spaceand image morphological processing, images are segmented and connected regionsare divided in different groups. Sub-pixel accuracy accurate extraction of the centercoordinates of the target signs are achieved by the two geometric roundnessconstraints and circular least squares fitting quickly and accurately. Contrasting to the Hough transform based circle recognition algorithm, experimental analysis of therecognition accuracy rate of the landmarks of passive cooperation objectives andprocessing time are implemented. Experimental results show that our proposedalgorithm is superior than the Hough transform based algorithm.Under the constraints of rigid target moving, the least squares and standardreconstruction equations of circle center of the passive cooperation objectives signsare given separately. The reconstruction uncertainty equation is derived by analyzingthe uncertainty description of linear and nonlinear models.A precision weight constrained target motion estimation algorithm is proposed.The motion equations contains an error term is derived using the reconstruction of thecenter of the circle signs and rigid body kinematics transform equations. To improvethe robustness of motion estimation algorithm, nonlinear weighted least squares areutilized to build objective functions with the non-normal distribution errors.According to the reconstruction uncertainty of the signs, the precision weightingfactors are computed to obtain a closest minimum unbiased estimation of targetmotion parameters by the nonlinear solution. To solve the rotation transformationmatrix orthogonal constraints, the objective function solution is anequality-constrained nonlinear optimization problem. Lagrange transform is used totransform it into an unconstrained nonlinear optimization problem. The solutionequations of the rotation transformation matrix and translation transformation vectorare derived using the characters of the symmetric matrix and singular valuedecomposition algorithm to avoid local minimum in non-linear iterative solution.