Object Detection With Sparse Representation Based Deformable Part Model

Object detection is extracted from the captured images of the area of interest, as a basic and important problem of image processing by the attention of scholars both at home and abroad, it has a wide range of applications in object detection, visual navigation, space remote sensing and so on. For the detection and location of one kind of object (such as a person or car) from the static images, due to the appearance of the object in these categories differ in thousands ways, so the work becomes very complicated. The object detection with Discriminatively Part Model (DPM) method is proposed by P. Felzenszwalb in 2008, is a robust and efficient method for object detection. At present, DPM has become the core part of numerous classification methods, segmentation methods, and pose estimation methods. The object detection with Discriminatively Part Model (DPM) method have achieved good results in recent sessions of PASCAL VOC Challenge.The object detection with Discriminatively Part Model (DPM) method uses histograms of oriented gradients (HOG) to describe features. HOG limits the performance of DPM, as it cannot deal with noisy edges and ignore the flat areas while focusing on edge areas. In recent years, the research on sparse representation is becoming more and more hot, especially in image processing and recognition. In order to improve the performance of DPM, an object detection with Sparse Representation based DPM is presented. Instead of using HOG, the method uses sparse coding to construct a new feature descriptor. The sparse coding based feature vectors can represent more information of image patches.The experimental results show that the method can improve the DPM method's precision on the PASCAL VOC 2012 dataset.