Based Moving Target Detection In The Monocular Visual Motion In The Context Of The Relevant Studies

The object detection based on video frequency is an important topic in computer vision. To separate the object from the real-time moving background rapidly and accurately, is the key to the further analysis with the video frequency. The object detection based on the immovable camera has been widely used in the fields of modern military and civilian. Now, the object detection based on the moving background has become new focus.The object detection technology and the background motion compensation technology based on single camera on the moving background were further studied in this paper. An improved RANSAC algorithm and LS algorithm based on feature blocks was proposed. The robust statistics algorithm was used to estimate the global motion in order to finish the background motion compensation.First, a kind of circle template was designed according to the basic theory and the moving background environment to improve the accuracy of blocks-matching and Pre-detection method was used to detect the feature blocks, which could reduce much time from matching the feature blocks. Second, the improved RANSAC (Random Sample Consensus) robust statistics was used to remove the blocks in the target area and those blocks which are matched falsely. We used a threshold selection method from gray-level histograms proposed by Otsu to choose the blocks that would be concerned with the parameter estimation of final model. Finally, the final blocks was used to computer the global motion estimation parameters by LS (least squares algorithm) to finish the background motion compensation.The experiments showed that the improved algorithm could get better effect of background motion compensation toward the translational and rotational motion of the background.In the process of object detection, the inter-frame difference object detection method and symmetrical inter-frame difference object detection method based on gray were used respectively. The experiments results proved the validity of background motion compensation algorithm and the feasibility of object-detecting based on the background motion compensation. Aiming at the case that target area was larger and the gray was distributed with less levels, a new algorithm on moving background was proposed to lessen the connectivity rupture and to minish the hole in the target area.