Object Detection And Tracking Algorithm Based On Blob

Moving object tracking technology is a hot research in the field of computer vision.In the early days,it is mainly used for military,medical,industrial and other specialized areas.Now it is gradually expended to consuming field,such as intelligent video surveillance,automotive navigation,interaction services and so on.There are many challenges in the moving object tracking task,mainly includes real-time requirement and robustness.This paper will attempt to implement a robust object detection and tracking algorithm.Blob is a connected component of the figure,it can describe a separate object.And the blob can be used in the object detection and tracking.When we obtain the blob of an object,we can calculate the color feature of this blob and use this feature to track object.Someone also combine the blob and Mean-shift tracking algorithm to adjust the object searching window.In this paper,the distance between blobs is used to match the object.There are three kinds of object detection algorithms,including frame difference,background subtraction and optical flow.In the background subtraction,there are many background model algorithms,like median model,average model,Gaussian Mixture Model and so on.The Gaussian Mixture Model can adapt the complex scene,so this paper will use it to maintain the background and detect the foreground.Then this paper will utilize the connected component extraction algorithm to extract the blobs and calculate size and centroid of these blobs.These two algorithms can realize the object detection.In the process of the object tracking,the distance between observation blob and tracking object will be used to solve the object matching problem.Firstly,a data association matrix between observation blob and tracking object should be built.Then this paper will use Munkres algorithm to solve this matrix.According to the result of data association,this paper will use the Kalman filter to manage and track the objects.Based on the realization of my algorithm,this paper have tested the algorithm according to the performance evaluation criteria about multiple objects tracking.And the test result has proved that my algorithm is reasonable and feasible.