Study On Video Object Detection And Matching In Video

With the rapid development of computer and digital image processing technology,intelligent monitoring based on computer vision has been gradually applied to public safety,intelligent transportation,military,industrial testing and other fields.While the detection and tracking of moving objects in video is one of the key problems in the research of intelligent monitoring.One of computer vision objectives is that the computer has a real "see" skill through dealing with video.In this paper,we study the homography estimation and the detection and tracking of moving objects in computer vision.The main work is as follows:(1)How to estimate accurately the homography is still a challenge problem in computer version.Until now,the homography estimation is almost under the RANSAC framework.In this paper,we explore a different approach,which is achieved by means of order-preserving constraint and the similarity measurement of the four points of the quadrilateral.The proposed method don't need under the RANSAC framework and don't need the repetitive the estimation of the homography.Thus it is less time consumption.In the synthetic experiments,we analyze the influence of the proportion of outliers and noise level for the estimation of the accuracy.We also compare the results of the proposed method with other methods in real image experiments.(2)Aiming at a few problems on the traditional TLD(Tracking-Learning-Detection)algorithm,such as time consumption,the tracking speed is slow,the moving target with high speed leads to track failed and so on.This paper improves the tracking module and the detection module in the TLD algorithm.In the improved method,the tracking and detection are implemented in the target neighborhood of previous frame in terms of the continuity of the target motion.The HOG feature detection and SVM classifiers are introduced into the detection module.Multi-core parallel programming technology is also adopted in order to further accelerate the speed of the algorithm and to achieve real-time results.In the experiments we verify the ability of the improved TLD algorithm to overcome the shielding,re-track again and track the moving target with high speed by detecting and tracking the soccer and player in a real soccer game video.The experimental results show that the tracking accuracy,real-time and robustness of the improved algorithm are better than the traditional TLD algorithm.