Twin network infrared target tracking method based on feature pyramid

Abstract

The invention discloses a twin network infrared target tracking method based on a feature pyramid. The twin network infrared target tracking method includes the steps: respectively carrying out full convolution operation from bottom to top on the template frame and the detection frame, then respectively carrying out top-to-top operation and transverse connection on convolution layers C5, C4, C3 and C2, and respectively and correspondingly generating P2, P3, P4 and P5 scale feature layers; according to the classification weight and the regression weight of each scale feature layer of the template frame, carrying out convolution operation on the corresponding scale feature layer of the detection frame to determine a scale proposal corresponding to each scale feature layer of the detection frame; and performing non-maximum suppression on the scale proposal corresponding to all the scale feature layers of the detection frame, retaining the highest scale proposal, and taking the scale feature layer corresponding to the highest scale proposal as an output tracking result. The twin network infrared target tracking method has better discrimination capability for target deformation, and ismore suitable for complex scenes and target deformation.

Description

technical field [0001] The invention belongs to the field of infrared target tracking, and in particular relates to a feature pyramid-based twin network infrared target tracking method. Background technique [0002] In recent years, moving target detection and tracking technology based on visible light computer vision has developed rapidly and has been widely used in various fields, such as human-computer interaction, intelligent video surveillance and precision guidance. However, the visible light system cannot play an effective role in bad weather and at night. Compared with it, the infrared imaging system has a relative advantage in this respect, because the infrared imaging system obtains information by sensing the infrared signal radiated by the object, so it has It has the advantages of working around the clock, good concealment, strong ability to penetrate smoke and dust, and strong anti-interference ability. [0003] Infrared target tracking is the process of findin...

AI Technical Summary

Problems solved by technology

Infrared images lack visual color information, low image signal-to-noise ratio, difficult to distinguish similar objects, serious background clutter and low resolution, making infrared targets easily submerged in the background

In addition, if there are similar objects in the background, the infrared tracker is very easy to follow the wrong target, and it is difficult to re-detect the target after it is occluded

For example, the twin network tracking algorithm (SiamFPN), which cannot output accurate target size, and is extremely error-prone when there are objects similar to the target in the background

Images