Visual frequency humary face tracking identification method based on appearance model

Abstract

Video frequency face tracking identify method based on appearance model belongs to video frequency dispose technology field. In the training video frequency, get person face through sample tracker and handy dispose and project training picture of each object to low-dimensional space by local linearity inlay technology. In the low-dimensional space, take means clustering to divide person faces into some species based on different gesture or expression. During each of group of pictures, take robust local reserved mapping to approximate the non-linear submanifold with linearity character space and learn the dynamic character. Carry the tracking identify of person faces in the testing video frequency. The tracking identify takes same appearance model and greatly enhances the tracking and identify capability of video frequency person face. It abroad uses in all kind of public residence and military system, such as the optic controlling, video frequency inspect controlling, video frequency meeting controlling, robot optic navigation system and military affairs goal tracking identify system.

Description

Technical field [0001] The invention relates to a method in the technical field of image processing, in particular to a video face tracking and recognition method based on an appearance model. Background technique [0002] Video face tracking and recognition has very important applications in many occasions, such as vision-based control, man-machine interface, access control, intelligent surveillance systems, etc. The uncertainty of tracking and recognition has always been a difficult problem in video face tracking and recognition systems. There are generally two types of tracking and identification systems: (1) Tracking first and then identifying. In this type, tracking and recognition are regarded as two completely independent modules, which use different models, and the system only completes tracking and recognition in a simple sense. In this type of system, the tracking result directly affects the accuracy of recognition, but the recognition cannot adversely affect the tracki...

AI Technical Summary

Problems solved by technology

However, before establishing the exterior submodel, the article only performs mean clustering in high-dimensional space, and uses a simple eigenface method to establish a linear subspace, which will result in the establishment of exterior manifolds that are not robust and cannot improve the recognition of special features. It is the robustness of the target in video recognition under the occurrence of expression, posture and illumination changes.

Images