Vision based state detection for AMR

Abstract

In recent years, automation has become increasingly popular in industry scenarios, and it has brought tremendous growth and innovation to the manufacturing industry. The project investigates the optimization and implementation of industrial solutions for automated transport processes. In this report, the author designs and implements a vision-based state detection system for AMR. This object detection system which is based on template matching contains functions including generating templates from video or separated images, the objective detection module based on the DCNN network, and the instant detection module operated by OpenCV. An optimization method based on object tracking called Sort is also implemented and adjusted for predicting the next frame bounding box and filtering the false detection in order to stabilize the detection system. In this project, the video is obtained from the camera and an instance detection system is used to detect AMR, including localizing the position and testing the detection status. After the object is detected in the camera, the system controls the opening and closing of the elevator door to achieve automatic transfer