Long-time tracking method for a mobile platform

Abstract

The invention discloses a long-time tracking method for a mobile platform. The method comprises the following steps: reading in a first frame of image of a video, selecting a target tracking box, andinitializing a tracker, a detector, a predictor and a detection mark; After determining a target tracking frame, reading the next frame of image, and tracking the target by using a tracker; Judging whether the tracking is successful or not, if the tracking is failed, re-detecting the target by using a detector, if the target is detected, calculating the central position and the outline frame of the target by using a detection result, and otherwise, performing predictive tracking; If the tracking is successful, calculating the central position of the target and the size of the outline frame byusing a tracker; According to the method, the actual mobile visual platform can track for a long time without losing a target, and the problems of shielding, target motion nonlinearity, noise non-gauss and the like which easily occur in the tracking process are solved.

Description

technical field [0001] The invention relates to the field of intelligent control, in particular to a method for long-time tracking of a mobile platform. Background technique [0002] In mobile platforms such as unmanned reconnaissance vehicles and drones that use visual sensors, intelligent tracking of specific targets has always been a research focus and difficulty. Most of the current methods consider the target tracking of the video in the surveillance scene, which is characterized by a fixed camera and only short-term tracking of the target. In the vision system application scenarios of unmanned reconnaissance vehicles and UAVs, the tracking of a specific target is not only long-term tracking, but also the image sensor needs to move through the gimbal to continue tracking the target when the target is about to leave the field of view. Images captured by image sensors are also more likely to be polluted by noise. The occlusion problem is often considered in existing met...

AI Technical Summary

Problems solved by technology

In the vision system application scenarios of unmanned reconnaissance vehicles and UAVs, the tracking of a specific target is not only long-term tracking, but also the image sensor needs to move through the gimbal to continue tracking the target when the target is about to leave the field of view. Images captured by image sensors are also more likely to be contaminated by noise

The occlusion problem is often considered in existing methods, but most of them are for short-term occlusions. The existing methods are only suitable for the situation where the target out of the occlusion is still within the video range, and is not suitable for the situation where the target is out of the field of view due to a large block of occlusion.

There are also existing methods that are only applicable to the situation where the target motion is linear and the noise conforms to the Gaussian distribution, which does not exist in the actual scene

Images