Abstract:

The traditional template-based matching algorithms generally consider the integrity of the template, which brings often the problem of computational complexity. This paper proposed a fast approach to texture-less object detection based on the binarized orientation compressed map and local feature weighting. According to the quantized gradient orientation of the edge points, the proposed method described the representation of the template features by using the orientation compressed map, which could estimate object candidate locations and corresponding scale and rotation information with fast-speed. After the detection of the object candidate locations, a new template feature was built based on the local feature weighting, and the novel similarity of the object candidate locations was calculated to determine the final pose of the object. Experiments showed that the proposed method had better recognition rate than other representative algorithms, and the recognition speed was greatly improved.

Key words: texture-less object, object detection, template matching, object candidate locations, the binarized orientation compressed map, the local feature weighting