Target tracking method and device

Abstract

The embodiment of the invention provides a target tracking method and device, which can generate a sampling frame in a Tth frame image acquired by a camera device, the sampling frame comprises a firstarea of interest of a target face, cyclic shift is performed on the sampling frame in the Tth frame image to obtain n training samples, each training sample corresponds to a classification label. A nonlinear classifier is trained by using the n training samples and the classification labels corresponding to the n training samples, the response degree of the input samples and the first area of interest is outputted by the nonlinear classifier, and cyclic shift is performed on the sampling frame in the (T + 1) th frame of image to obtain n shift samples, A target sample is determined with the highest response degree with the first area of interest from the n shift samples by using a nonlinear classifier, the first area of interest is updated according to the target sample, and the visual angle of the camera device is adjusted according to the position of the first area of interest. The technical scheme can adapt to scenes such as multiple faces and high dynamics in the visual field of the camera device, and is high in robustness.

Description

technical field [0001] The present application relates to the technical field of computer vision, and in particular to a method and device for object tracking. Background technique [0002] Robots can generally refer to man-made devices that have the ability to work semi-autonomously or fully autonomously and complete human work, such as household robots for personal or family use, food delivery robots in hotels, and guidance robots in hospitals and other places. Welcome robots in bank halls and security robots in public places, etc. [0003] The robot uses its camera device to collect images of the user, judges the user's position, and adjusts the direction of the camera device according to the user's position, so that the camera device always faces the user when interacting with the user, and realizes target tracking for the user. However, the current method of object tracking based on the camera device still has the problem of poor robustness in practical applications. ...

AI Technical Summary

Problems solved by technology

However, the current methods of target tracking based on camera devices still have poor robustness in practical applications.

For example: when many people enter the field of view of the camera device, the robot will detect multiple faces, so that it cannot distinguish which face is the face to be tracked, resulting in the failure of target tracking; in addition, when the target has only part of the body In the field of view of the camera device or when the target is facing away from the robot or facing the robot sideways, the robot may not be able to detect the human face, resulting in the failure of target tracking, which may cause the camera device to be unable to track the target; When occluded, the robot will directly track the non-target face, resulting in target tracking failure

Images