Machine Vision-based Recognition and Positioning System for Domestic Garbage

Domestic garbage classification is a global issue, and the robotic arm grasping system for garbage sorting can greatly improve sorting efficiency. Such robotic systems recognize and locate garbage by machine vision. The problems of these existing systems are as follows: the recognition rate for destroyed and part-missing garbage is low; garbage positioning is inaccurate and time-consuming when the environment is dim or light-reflection. To this end, the paper collects 3116 images of domestic garbage, and divides them into 4 major categories and 12 minor categories; and then proposes an improved CenterNet target detection algorithm which is combined with the convolutional attention and feature fusing; normalized cross correlation matching algorithm has been reformed to enhance the positioning accuracy in the light-reflection environment. In addition, pre-cutting and pre-matching are executed to improve object positioning. The experiments shown in this paper prove that the system can recognize domestic garbage which is destroyed and part-missing and locate it in the dim and light-reflection environment.