Object detection algorithm based on deformable part models

The paper proposes an object detection algorithm based on the deformable part models, and integrates the idea of global and local information to improve the accuracy and robustness of target detection. Firstly, we train the pyramid HOG feature of the sample images and get the feature representation containing the root model, component model and the corresponding deformable part models, then use the HOG features to train the classifier LSVM. Finally, we use the algorithm of dynamic programming combined distance transformation to section out the region on the detected images that matches the deformable part model, thus achieve the location of our interested target. The experimental analysis indicates that the proposed method can solve the problem of localization when the targets are blocked or interfered in the complex environment.