Unmanned aerial vehicle aerial video moving small target real-time detection and tracking method

Abstract

The invention discloses an unmanned aerial vehicle aerial video moving small target real-time detection and tracking method. The method comprises the steps of obtaining a current background model through modeling of a single Gaussian background model, distinguishing whether pixel points are foreground or background after obtaining the single Gaussian background model so as to obtain a foreground image, performing sparse optical flow analysis on the obtained foreground image, and obtaining a tracking point set; Hierarchical clustering is carried out on the tracking points; obtaining an outer frame of the tracking target, deducting a tracking target in an outer frame of the tracking target obtained by detecting each frame of foreground image; a follow-up to-be-tracked list is formed, a feature vector is extracted from each tracked target through a deep neural network, prediction is conducted on each tracked target through a Kalman filtering algorithm, matching is conducted through a Hungarian algorithm, the tracked list is updated, and the updated tracked targets are obtained. According to the invention, a single Gaussian model is adopted to carry out background modeling, the detection time consumption is reduced, and the overall efficiency is improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a real-time detection and tracking method for a small moving target in an aerial video of an unmanned aerial vehicle in the technical field of target detection and tracking. Background technique [0002] Moving target detection and tracking tasks in UAV scenes face the same problems as occlusion, shadows, and environmental interference in other common video scenes, but they also have unique characteristics. For example, the weight of the UAV is limited, and it is necessary to achieve near real-time detection and tracking effects under the condition of limited computing resources. In addition, in the UAV video surveillance scene, not only the target to be detected is moving, but also the camera is moving, which is easy to cause confusion between the foreground and the background. Moreover, due to factors such as camera movement and weather changes, lighting co...

AI Technical Summary

Problems solved by technology

However, the accuracy and efficiency of the current tracking methods need to be further improved, and they are not robust enough to target occlusion and light transformation.

Images