Multi-Target Vision Detection and Grasping of Electronic Devices Using DL-SORT

Abstract

Many production industries are increasingly dependent on intelligent electronic devices. However, traditional methods for recognizing and detecting electronic devices are inefficient and consume a large amount of production resources. This study introduces a hybrid DL-SORT model for automatic recognition and grasping of electronic devices, integrating deep learning with Simple Online and Realtime Tracking (SORT) to enhance object detection performance. In the model, Recursive labeling and Binary Robust Independent Elementary Features are also employed for key point detection and domain selection. Experimental results show that the hybrid algorithm outperforms Single Shot MultiBox Detector and Discriminative Correlation Filter with Channel and Spatial Reliability in terms of object detection performance, with loss values of 0.15, 0.24, and 0.23, respectively. Additionally, empirical analysis of the constructed hybrid model reveals that the proposed automatic recognition and grasping model for electronic devices achieves an accuracy of 0.95 in tape recognition, demonstrating good recognition accuracy. Testing in obstructed environments shows that the success rate of part detection remains above 80%, with minimal performance degradation. These results suggest that the hybrid model can detect multiple targets and improve production efficiency. This study contributes to the future development of drones and industrial robots in the automation field, enabling the acquisition of precise target location information.