Distribution-Matching Stack Object Counting Based on Depth Information

Abstract

Object counting is one of the crucial research fields of computer vision. Most of the existing counting methods are based on the plane counting of a single image. In the application field of industrial intelligent management, the counting of materials is very imperative, but the materials are usually placed in a stacked manner, and the existing counting technology is unintelligible to solve this problem. This paper proposes a distribution matching planar object counting method based on depth information. While optimizing the planar density map, depth information is introduced for the first time to assist in counting stacked objects. Experimental results show that our method remarkably reduces the error in the generation of planar density maps, and is effective in solving the problem of counting stacked objects.