Study On Human Face Images Acquisition Technology Under The Unconstrained Conditions

In recent years, with the rapid development of the application of the biometric identification technology in the safety-security area, the face recognition technology has also attracted more and more attentions as an important technology of the biometric identification technology, and is gradually used in the fields such as entrance guard, security inspection, smart space and natural human-computer interaction. Under the constrained conditions, the recognition rate can reach more than 95% (FERET database), but under the unconstrained conditions, the recognition rate will decrease rapidly and can't meet the actual applicative demands.This paper first analyzes the recent research results of the face recognition technologies under the unconstrained conditions, then discusses on the ways to improve the recognition rate under the unconstrained conditions and points out that improving the face images acquisition technology is a useful method. For the image understanding is an important branch of the artificial intelligence, the face images acquisition technology researched in this paper is under the human visual principle. In order to acquire a human target's face image, the human should first have a large-scale scanning in the "where" vision to acquire the location information of this human target, then rolls their eyes to the location and acquires the face image in the "what" vision. According to the principle, a multi-vision-sensor fusion device is designed in this paper to simulate the face images acquisition process of the human: In the "where" vision, the Omni-Directional Vision Sensor (ODVS) is used to acquire the panoramic images of the monitoring scene, and the human detection and tracking algorithm is used to acquire the location information of every human target in the monitoring scene. After acquiring the human location information, the head location information can be calculated, and then the PTZ cameras in different azimuth can be controlled to roll to the head location and acquire the head images of the human target in the "what" vision. Finally, the images that can be used in recognition will be filtered out according to the facial feature information detected by the face detection algorithm and eye detection algorithm. If no image is filtered out, the PTZ cameras will be controlled to acquire the head images of the human target again to ensure that the face images that can be used in recognition are acquired.Besides, a human face images acquisition system based on the face images acquisition device is also developed in this paper, and the functions and implements of every module of the system are also introduced in this paper. The experimental results show that this device and system can well acquire the images that can be used in recognition, and have a good theoretical significance and practical value.