Detection And Tracking Of Moving Object In Video Image Sequences

Video and image processing techniques are widely applicable to industry, medical, security and management. The detection and tracking of moving objects, as a part of image processing techniques involved in video image sequence processing, image processing play an important role in several fields, such as aircraft, traffic, robot vision, video surveillance and pedestrian flow data in public establishments.In this paper, based on the study of algorithms for detection and tracking of moving objects and on the analyzing of problems in applications, improved methods are proposed to tackle with the problems. And the experiment results proved their efficiency and stability.The motivation of detection of moving objects is to detect the pixels that belong to moving objects area in this frame image. Traditional methods of detection contain background distraction, lighting flow and frame distraction. Frame distraction detects the moving objects pixels by the difference in gray value between the current frame image and its adjacent frame image. Experiment result of Traditional two frame distraction may be out of shape, stretched and with several large area empties. Three frames distraction may not get the clear edge of the moving objects. Considering with these problems, this paper present an improved method of odd number frames distraction to obtain relatively accurate shape, more edge pixel and less empties. The result matrix, calculated through the detection method, should be converted to a binary image with the appropriate threshold. After the image morphology processing, this image can be the result of detection. On condition of the accurate detecting result, the accuracy of feature extraction in the next step of processing about tracking of the moving objects can be promised.Tracking of moving objects is the prediction and estimation of the object position, or to match the feature of the current frame image with the template in the searching window. Tracking algorithm is based on the extraction of the object feature in several types, such as position, shape and color. Take Kalman filter tracking and mean shift tracking as the examples of widely used tracking methods. According to the moving object area by the detection, the center position of the object can be figured out and act as the observation vector. Considering the circumstance in the image scene, the state function and observation function can be constructed. The predictive state value of the current frame image is the result of object tracking. Count the number of every single value of hue in the HSV space, construct the statistics histogram and match the histogram with the one of template. By maximization of the Bhattacharyya coefficient, the most matching position can be obtained. The right position in the searching widow is the result of mean shift tracking. Kalman filter tracking use only one feature of position, while lighting and velocity of object have an infection on mean shift tracking. Based on the analyzing of the defects and problems, this paper propose an improved tracking method that makes the use of object features and is capable of updating the template in the mean shift algorithm. The method gets the sum of weighted results of the two methods above and adjusts the weights to the objective conditions.