Detection And Tracking Of Camouflaged Moving Object Under The Dynamic Background

Moving object detection and tracking is an important research topic in the field ofmachine vision, image processing and pattern recognition, its main objective is to obtain theposition and motion features of the moving object. It’s one key underlying technology but nota separate research module, because its process results can directly affect the furtherprocessing and analysis and it also affect the higher level tasks. For the detection and trackingof moving object, scholars have proposed their own algorithms, but the detailed research onthe detection and tracking of camouflage object haven’t been done, especially on the trackingof camouflage object. The camouflage object is so similar with the background of videosequences that those traditional detection and tracking algorithms can’t be simple for solvingsuch moving object. In this paper we carried out the basic research on the camouflage object.In this paper, we took a comprehensive analysis on camouflage object to determine themost unusual feature between it and the background was its motion feature, so the detectionand tracking work of camouflage object were carried out based on its motion feature.In the moving camouflage object detection module, a method based on motion featureand topological feature was proposed to detect the moving camouflage object. Firstly, themoving optical flow field of video sequences was calculated based on multi-resolution opticalflow, and then the initial segmentation was completed used the k-means algorithm based onobject’s movement consistency. Because the stable and effective object area couldn’t beobtained based on optical flow, the topology consistency of area was added to determine theposition of the object. The experimental results demonstrated the proposed approach waseffective to detect the relatively prominent camouflaged moving object with significanttranslational motion compared with the changing background.In the moving camouflage object tracking module, two tracking methods for camouflageobject were proposed on the basic of the completed detection. One method was based onbipartite matching, firstly the main motion direction and speed size were calculated based onmulti-resolution optical flow, at the same time the hog feature of object area was also obtained,then the initial object area was bipartite matched in the range of speed size along the maindirection of object. This method could reduce the match times and improve the track efficiency. The experimental results demonstrated the proposed approach was effective totrack the camouflaged moving object in slowly changing background when the color ofcamouflaged object changed greatly or the moving object had obvious edge structure. Theother method used kalman tracking filter algorithm to do a simple tracking for movingcamouflage object.In this paper, the basic research work for the detection and tracking of movingcamouflage object was given, the further research under the background with changingviolently has to be carried out to find one a strong robust and widely used method forcamouflage object.