MOTION OBJECT RECOGNITION IN SURVEILLANCE VIDEO BASED ON FUZZY SUPPORT-VECTOR-MACHINE

摘要

A method for recognizing motion objects in surveillance video is presented. Firstly, according to fuzzy theory, a fuzzy feature vectors (FFVs) extracting algorithm is presented, which describes the object features by membership degree. Finally, with a "one-against-rest" classifying strategy, the multi-class pattern recognizing problem of motion objects is solved by using a multi-level two-class support-vector-machine (SVM) classifier to classify and recognize. So, motion objects in videos are classified into four patterns which are named as "people", "vehicle", "motorcycle", and "bicycle with man". The experimental results show that the method presented in this paper is feasible and good. It can correctly recognize each motion object even in the situation of the object has been scaled and distorted.