Abstract:

Aiming at the complexity of articulated object detection, this paper proposes a novel casting of visual inference, based on a standard object model, makes full use of the edge feature of the image(it does not rely on local features), as well as the space information of objects, and performs an iterative parsing process, where we can sequentially handle the articulated object detection and estimate the local object components. Experiments show that this method has a strong resistance to the background interference, and can also greatly improve the results of articulated object detection by human vision.

Key words: object model;edge feature;iterative;object detection;localization