A Vision-based Dual-axis Positioning System with YOLOv4 and Improved Genetic Algorithms

Shan-Ling Chen,Shang-Chih Lin,Yennun Huang,Chia-Wei Jen,Zheng-Long Lin,Shun-Feng Su

DOI: https://doi.org/10.1109/irc.2020.00027

2020-11-01

Abstract:This research aims to construct a vision-based dual-axis positioning system and make quality inspection more automated. The main tasks of the system are object detection and path planning. First, the latest object detection algorithms are applied to metal object detection and the definition of feasible inspection regions. Then, any object may have multiple feasible inspection regions, and the center point of each feasible inspection regions becomes the target of the path planning algorithm. The experimental results show that You Only Look Once (YOLO) and Improved Genetic Algorithms (IGA) realize object detection and path planning respectively, and have the best performance. The former will not be affected by the angle and distance of the object. The latter used optimization methods to reduce computational costs and optimal path planning. In future research, we will expand the scope of research. Let the algorithm handle more complex situations, such as considering the in-depth information of metal objects and performing quality inspection based on deep learning. In this way, the results of this article can help industrial technology upgrade and move towards smart manufacturing.