关键词:  机器视觉； Fisher判别准则； 动态多目标识别； 目标分割

Abstract:  This paper proposes a realtime dynamic multitarget recognition method based on machine vision. The method first segments the moving targets according to the pixel change between the front and rear frames, then extracts the features of the segmented moving targets and sample images using Gabor filter to obtain the eigenvector, and finally uses Fisher discriminant criterion for classification and recognition so as to automatically lable the classification and recognition results onto the output images. The images labeled with the recognition results are to be outputted continuously, and hence the output videos which have been recognized can be obtained. The experimental results show that in the circumstance of multiple targets, the proposed dynamic target recognition method integrating Gabor feature and interframe difference can accurately detect the dynamic target area, and realize classification, recognition and labeling.

Key words:  machine vision; Fisher discriminant criterion; dynamic multitarget recognition; object segmentation