Electric Power Monitor tracking algorithm based on Improved SiamFC

With the deep application of intelligent safety monitoring technology in the field of power engineering construction, computer vision technology represented by object tracking has become one of the important research contents in the field of smart grid. In the industrial visual recognition and early warning system, it is difficult to achieve accurate and real-time tracking. In this study, we propose a real-time target tracking algorithm based on improved SiamFC. The second convolution layer in the original siamese network structure is replaced by deeply separable convolution, improves the tracking speed by reducing parameter calculation, meet the need of real-time tracking in practical application. In the third convolution layer, mixed deep convolution is used to extract features through convolution kernels of different dimensions to achieve multi-feature fusion, extract features with stronger robustness, and improve the network's ability to distinguish objects and backgrounds. The performance of the algorithm is tested on OTB2015 data set and power construction site surveillance video. Experimental results show that compared with the SiamFC algorithm, the algorithm has a certain improvement in the tracking success rate, tracking accuracy and tracking speed, and can meet the tracking requirements in power construction scenarios.