Video-based Moving Target Detection And Tracking

In real life,large amount of meaningful visual information is included in the campaign. While humans can see both dynamic and static objects,but on many occasions,for instance, the security supervision for important place,industrial processes' dynamic monitoring, dynamic robot vision,automatic target tracking,automatic self-piloting and assistance vehicle driving,traffic flow control,the guidance of aviation and military aircraft,in which cases people are often only interested in the moving goal or the objects.Therefore,there will be great practical significance and practical value on the research of the moving target detection and tracking.The selection of this paper is based on one of projects that belongs the Ministry of Education" the parental affection plan".The name of item:the face recognition applied in family robot.According to the division of the topic,I am responsible for the picture detection and tracking in moving target.With the possibility that robots enter the human's family life increasing,How to solve the exchange problems between the robots and human is becoming an urgent issue.When the robot has the demand of exchanging,we need to lock the target of the exchange,and also we need to know which object(object code)is in the movement as well as the movement information(position,speed and direction)when the lock-on object campaigns.For the reason that the robot must make a corresponding position adjustment according to above information.Moving target detection and tracking is such a process that let the computer determine the target current position,movement parameters,spatial structure,and give the corresponding tracking movements according the calculated similarity between the image obtained by the sensor(real-time)and the reference image which contains the object.This paper has realized the moving target identification and tracking systems which is under the daily life environment.Mainly discussing the situation in which multi-moving goals are automatic tracked.While there are many moving objects in the field of view,in order to improve the image processing speed and real-time,we first carry on gray scale transformation to the current image and update the background image for real-time through Surendra background updating algorithm to obtain a background image.In accordance with background subtraction the foreground image can be obtained,also threshold can be accessed by the method OTSU(maximum between-class variance thresholding method).Taking binary image processing to the gray sequence,which can highlighting the targets in image.Based on the above Theory and combined mathematical morphology and connectivity testing,we can eliminate the noise disturbance and separate the regional of moving objectives,at the same time update background and get the location and size of moving target.In the tracking process,Mean Shift algorithm is first used.But it is regard to be lack of real-time as it needs to calculate every pixel.The Kalman filter can make historic filter of the location of targets and get the parameter estimating,which can significantly reduce the search scope of the template matching.Combing the advantages of these two algorithm,this paper used the method that joins Mean Shift algorithm and Kalman filter to track the moving targets,which comes to the to requirements that tracking and dealing with the targets rapidly and steadily.According to OpenCV(Intel @ Open Source Computer Vision Library),meanwhile based on Mean Shift algorithm and the Kalman filter,the automatic tracking of the color image sequence is achieved.Finally the current system and the existing shortage are analysed and the improvement can be done in the future is also proposed.