Yan Liu,∗ Binhu Wang,∗ and Jiapeng Yang∗
Machine vision, industrial robots, improved YOLOv4, product identification; product grasping
As industrial automation technology develops rapidly, industrial robots that integrate machine vision and robotic arm grasping have great potential in improving production efficiency and reducing labor costs. A deep separable convolutional layer is used to replace the standard convolutional layer of deep networks to improve YOLOv4 to improve the target accuracy of industrial robot product recognition and grasping in complex environments. Meanwhile, K-means++ clustering method is introduced in the improved YOLOv4 to intelligently select clustering centres and improve target detection performance. Moreover, a binocular vision system is used to enhance the recognition ability of small and partially occluded objects. These results confirmed that in both occluded and unobstructed environments, the improved YOLOv4 achieved the highest recognition accuracy of 93.64% and 96.91%, respectively, and positioning errors were reduced by 51.72% and 71.25%, respectively. The research on industrial robot product recognition and grasping based on improved YOLOv4 confirmed the effectiveness and accuracy of the improved YOLOv4 in both unobstructed and occluded scenarios. It shows better stability and robustness than the original model.
Important Links:
Go Back
