The Detection And Match Of Object For Multiple Non-overlapping Cameras

In recent years, with the rapid development of intelligent video surveillance system,as well as the urgent demand of the security situation, people’s requirements for smartintelligent video surveillance system are increasing more and more, and it is muchaccounted of day by day that multi-cameras network because of its expansive surveillanceareas. Nevertheless, on the one hand, due to the the consideration of capital and resources,it is unlikely to cover all surveillance area with a great amount of monitors. Therefore,camera tracking with non-overlapping views has become a main part in the study ofwide-area video surveillance. On the other hand, the precondition in the camera tracking isobject matching which is the significant research project in the object detection andmatching for multiple non-overlapping surveillance cameras.This thesis contraposes the objection detection of disturbed irregular background formultiple non-overlapping surveillance cameras, and the same target matching on multiplecameras, and with the in-depth study, the following results can be achieved:1.To the question the high rate of false drop of moving target detection in thedisturbed irregular background, based on distorted background difference model, replacingEuclidean distance with Bhattacharyya Distance to calculate the similarity between colorpixels with, we could get similarity of color pixels accurately. The background is dividedinto dynamic module and static module; utilizing the main color of the pixel module, thesimilarity discrimination of pixel modules can be done. When there is a lot of dynamicdisturbance, it adopts module distortion difference operation or distortion differenceoperation among pixel layer. In the aspects of recall ratio and precision rate, we comparethe detection of the moving birds in the forests with serious irregular disturbancebackground to the usual moving object detection, the experimental results show that it isvery effective that the method detailed in this paper that object detection for irregulardisturbance background.2.With obtaining all the targets’ Robust feature consistently with single camera view, the moving objection can be recorded uninterruptedly.it affects the tracking accuracy of thealgorithm, even worse, causes the failure of particle filter so that unable to complete thetracking problems. For this problem, the article proposes an improved pedestrians trackingmethod on the basis of the HOG and color features. The proposed algorithm uses theQuasi-Monte Carlo (QMC) sampling methods to get the more uniform distribution ofparticles instead of the MC sample set of particles, avoiding the gap and cluster problembetween the particles, and improving the accuracy of the tracking algorithm.improve theaccuracy of the target in the LDA match.3.To the question of accuracy of moving goal description in performance model formultiple non-overlapping surveillance cameras, we put forward a performance model ofmoving goal description based on region-SIFT descriptor When the target is in the range ofthe monitor, with Foreground Detection and Particle filter tracking, we could getregion-SIFT descriptors consistently with which we can build the performance model. Tomake region-SIFT descriptor correspond to moving goal, the matching work in the LDAmodel can be done. Experimental results show that the method is effective.4.To the question of the highly accurate rate of matching of object matching formultiple non-overlapping surveillance cameras with the sea change of moving object. Withthe SIFT characteristics of moving target, we could establish performance model torepresent those targets. Proceeding objection matching under LDA Model withperformance model established by all the SIFT characteristics of the objective area, In acertain extent, reduce the matching errors brought by the illumination change with thesimilarity gained from the performance model. Through experiment, it proves that LDAcan realize the model can object matching for multiple non-overlapping surveillancecameras effectively.