A real-time UAV video target detection and tracking method

Abstract

The invention discloses a calculation method for real-time detection and tracking of a specific target (mainly for moving targets such as tanks, cars, ships, etc.) in unmanned aerial vehicle video, which involves computer vision technology, machine learning technology and photogrammetry. On the basis of learning the existing targets, it detects the target in the current video frame and matches it with the target in the previous frame, updates the successfully matched target, tracks the old target, and confirms the newly added target. Finally, the target set of the current frame is obtained, and the target geographic location can be calculated and output according to the coordinates of the target in the camera. In the application of unmanned aerial vehicles, the present invention can discover in real time and automatically track and precisely locate the enemy's military-valued targets, such as tanks, cars, and ships.

Description

technical field [0001] The invention relates to the technical field of target detection and tracking in computer vision. To be precise, the present invention is a method for automatically finding and tracking a specific target from an imaging device during the flight of a drone, and involves technical fields including computer vision, machine learning and photogrammetry. Background technique [0002] The detection and tracking of specific targets is the key technology of digital video analysis and machine vision. , oil depots, etc.) for detection, tracking and positioning, and can return target location information in real time to facilitate subsequent precise strikes. It can not only reduce the labor intensity of the operator and enhance the reliability of the system; it can also reduce the communication with the operator, preventing the signal from being detected and causing the operator to be exposed. [0003] Video target detection methods are mainly based on scenes wh...

AI Technical Summary

Problems solved by technology

However, it needs to deduce the mean shift variable from the analytical form, which limits its feature extraction and has certain limitations; the tracking algorithm based on probability statistics has a good performance in solving nonlinear problems, but its equally complicated calculation affects the The tracking speed is poor in real-time; the learning-based tracking algorithm generally does not require complex calculations and has good adaptability, but it still has drift problems caused by template updates during the tracking process

Moreover, none of the existing algorithms can fully solve the problem of long-term tracking of UAV targets.

Images