Machine Motion Trajectory Detection Based on Siamese Graph-Attention Adaptive Network

There are many demands for motion trajectory detection of machines in intelligent manufacturing. The existing detection methods may lead to errors or loss of target motion trajectory. An anti-interference machine motion trajectory detection method based on the Siamese graph-attention adaptive network (SiamGAAN) is proposed in this article. A graph attention-based feature matching method is proposed in the feature extraction network of SiamGAAN, which transfers the template feature information to the search feature instead of the traditional cross correlation method. The regression network based on an adaptive anchor-free mechanism can directly predict the distance between the sampling points and the boundary of the target region. A quality determination-based index is proposed in the classification network to improve the accuracy of the target box. SiamGAAN can continuously detect the motion and collect the tracking information of machines in the complex environment of the workshop. SiamGAAN can achieve a 1.5% accuracy rate and 1.9% tracking success rate improvement compared with those typical methods on the benchmark data, i.e., OTB2015. A tracking success rate of 90.58% is achieved in the robotic arm's motion trajectory that simulates the workshop's real operation.