2011/12/06 by Duc Phu Chau, Chau, Duc Phu, François Brémond +3
Computer Science · #Advanced Image and Video Retrieval Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Human Pose and Action Recognition #Video Surveillance and Tracking Methods
paper · pdf · doi:10.48550/arxiv.1112.1200
openalex publication_date 2011/12/06 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We propose in this paper a tracking algorithm which is able to adapt itself\nto different scene contexts. A feature pool is used to compute the matching\nscore between two detected objects. This feature pool includes 2D, 3D\ndisplacement distances, 2D sizes, color histogram, histogram of oriented\ngradient (HOG), color covariance and dominant color. An offline learning\nprocess is proposed to search for useful features and to estimate their weights\nfor each context. In the online tracking process, a temporal window is defined\nto establish the links between the detected objects. This enables to find the\nobject trajectories even if the objects are misdetected in some frames. A\ntrajectory filter is proposed to remove noisy trajectories. Experimentation on\ndifferent contexts is shown. The proposed tracker has been tested in videos\nbelonging to three public datasets and to the Caretaker European project. The\nexperimental results prove the effect of the proposed feature weight learning,\nand the robustness of the proposed tracker compared to some methods in the\nstate of the art. The contributions of our approach over the state of the art\ntrackers are: (i) a robust tracking algorithm based on a feature pool, (ii) a\nsupervised learning scheme to learn feature weights for each context, (iii) a\nnew method to quantify the reliability of HOG descriptor, (iv) a combination of\ncolor covariance and dominant color features with spatial pyramid distance to\nmanage the case of object occlusion.\n