Intelligent substation human body target tracking method based on adaptive feature fusion

Abstract

The invention provides an intelligent substation human body target tracking method based on adaptive feature fusion. According to the method, while an accurate tracking result is rapidly given, a fine mask can be segmented from a target object, feature maps of all levels of the FPN are sent to an ASF of a multi-ratio anchor frame, an ASF output feature map is enhanced through a Gaussian non-local attention mechanism, target area features are highlighted, non-target and background features are inhibited, and the detection precision is effectively improved.

Description

technical field [0001] The invention relates to the field of target tracking, in particular to a human target tracking method in an intelligent substation with adaptive feature fusion. Background technique [0002] Nowadays, many positioning and tracking systems use ultra-wideband (UW) technology to complete personnel positioning and tracking based on hardware technology, but this method is costly and requires the monitored object to always carry a signal generator such as a bracelet, which cannot Track trespassers or construction vehicles. In terms of visual algorithms, some monitoring systems use traditional tracking algorithms such as adaptive mean shift (Camshift), support vector machine (SVM), adaptive enhancement (AdaBoost) or background extraction (ViBe), and Kalman filter to personnel or engineering vehicles to track. However, the traditional algorithm often relies on the similarity of the underlying features of the image. Although the speed is fast, it cannot adap...

AI Technical Summary

Problems solved by technology

However, the traditional algorithm often relies on the similarity of the underlying features of the image. Although the speed is fast, it cannot adapt to the large pose changes of the tracking target and the blurring and occlusion caused by the camera shooting, and it is difficult to output reliable images in more complex scenes. result

[0003] The substation target tracking method in the prior art mainly has the following deficiencies: (1) The tracking target of substation video surveillance is generally the staff in the station, and there is also a certain demand for engineering vehicles or intruding foreign objects in specific occasions

(2) The substation monitoring video has a low frame rate and low definition, and it is easy for the tracking target to be blocked by the equipment, which poses a great challenge to the intelligent monitoring algorithm

Images