Research On Multiple Moving Objects Recognition Technology Based On Video

Multi-moving objects recognition is an important part of intelligent video surveillance, and is also a hotspot in the area of computer vision. It has very important theoretical significance and value to do research on it. This paper uses moving vehicle, bike/motor and pedestrian in video as specific research objects.First, this paper adopts a strategy of median synthesis to establish an initial background model; segments the moving objects effectively by combining the self-adaptive background update algorithm with the weighted enhancing mean threshold method, and chooses area, shape complexity, length-width ratio, velocity as recognition features which have good distinguishability.Second, this paper does research on multi-moving objects recognition based on fuzzy theory. Fuzzy c-means clustering algorithm is adopted to realize features fuzzification. We designs Mamdani fuzzy classfier and Sugeno fuzzy classfier respectively. Simulation results show that Sugeno fuzzy classfier has better accuracy.Third, this paper does research on multi-moving objects recognition based on BP neural network. Aiming the insufficiency of BP neural network, the resilient BP algorithm and L-M algorithm are presented. We designs multi-output BP classfier and single output BP combination classfier respectively. Simulation results show that the last one has higher precision.At last, this paper does research on multi-moving objects recognition based on adaptive-network-based fuzzy inference system (ANFIS). Grid method and subtractive clustering method are respectively adopted to generate the initial structure of ANFIS classfiers. Simulation results show that the training speed of ANFIS classfiers based on subtractive clustering method is faster, and its precision is also higher. Integrating the merits of fuzzy system and neural network, ANFIS has transparent structure and self-adaptive learning ability. Its performance is the best.