Aircraft detecting and tracking method utilizing attention mechanism

Abstract

The invention relates to an aircraft detecting and tracking method utilizing an attention mechanism. A target detection algorithm based on a convolutional neural network is used to carry out target detection. The method is characterized in that a local attention enhancement mechanism is introduced; a large target convolution filter and a small target convolution filter are adopted to respectivelyextract large target feature information related to a target population and small target feature information related to a target specific part; and corresponding local information enhancement is carried out on the large target feature information by using the small target feature information, and target detection is carried out on the current frame image by using the large target feature information after local information enhancement to form a large target detection proposal. According to the method, background interference is reduced, the performance of aircraft detection is improved, tracker failure caused by long-time local shielding can be avoided, and the method is mainly used for aircraft detection and tracking in a flying area.

Description

technical field [0001] The invention relates to an aircraft detection and tracking method using an attention mechanism, which belongs to the technical field of computer vision. Background technique [0002] In the operation control of aircraft in the flight area, it is necessary to detect and track the aircraft in real time. Traditional aircraft detection and tracking is based on field surveillance radar. In recent years, the air traffic control operation system based on the panoramic video of the flight area has been actively promoted because it is more intuitive. [0003] Detection and tracking of moving objects in video are two important topics in the field of computer vision. In many application scenarios, the two are linked together. For example, Deep Sort algorithm [1] It is a typical method of tracking after detection. [0004] For target detection, the target detection algorithm based on convolutional neural network design is the current mainstream method, and th...

AI Technical Summary

Problems solved by technology

[0005] However, it is difficult to achieve high accuracy if the above-mentioned moving target detection and tracking methods are directly applied to the aircraft in the flight area (or called the airport)

There are mainly the following reasons: First, the target detection algorithm based on the convolutional neural network is not friendly to the detection task of the aircraft. This is because the area ratio of the aircraft in the aircraft rectangular frame learned by the detection model is usually not high enough. The features extracted by the feature extractor are caused by strong background characteristics; secondly, in the ground environment of the airport, it is normal for aircraft to have partial occlusion events, and the lack of specific local features of aircraft targets caused by partial occlusion may also cause target detection algorithms Failed to detect it; finally, there is almost no difference in appearance between aircraft of the same model (unless it is painted by an airline in a large area, but the paint of aircraft of the same airline is usually similar), and the appearance of aircraft of different models is different It is also small, which makes it difficult to distinguish different aircraft using image features based on the appearance of the aircraft, which limits the performance of the Re-identification function in tracking tasks

Images