Video Of Multiple Moving Targets Detection And Statistics

In modern world, the moving targets identification technique is always an active field in the intelligent visual system study. The moving object detection and flow analysis has a wide application need, obvious social and economic benefits in the city traffic environment. As the video image processing technology is rapidly developing, to accurately identify the different kinds of moving targets in the road traffic will be an inevitable trend.In road scenes, the most common objects are pedestrians, cyclist the vehicles. The research for the cycling pedestrian recognition is rare in papers at home and abroad. Because of the similarity of pedestrians and cycling pedestrians, how to effectively distinguish with the pedestrians is a great challenge.The AdaBoost algorithm was used in a lot of target recognition system. It has high recognition and robustness for the color, size, location of the moving targets and the outside environment changes. In these years, it was widely used in target recognition.A multi-target recognition method bases on the AdaBoost algorithm is proposed in the paper. An expanded descriptive feature from texture energy feathers is created by analyzing texture features of samples. These features are used to classify pedestrians, cycling pedestrians and the vehicles in video. Experimental results show the effectiveness of this method.This paper mainly researched selections and constructions of descriptive features based on the texture energy, not the AdaBoost algorithm itself. In order to filter texture features which have clear distinction with samples, we re-structure the contrast for the samples, and then obtain the texture energy values of the samples. According to energy distribution of different moving objects to construct the "日" sample fonts feature, "T" type feature, inflexion point type feature and linear characteristics, etc. The gray image was replaced by the texture energy image as the training samples. The classifiers were constructed by replacing the pixel values by the energy values. The test sample library was experimented, and the experimental results were compared and analyzed, the feasibility of the proposal in multi-target recognition was validated. The method had good recognition rate and robustness. Keywords:multi-objections recognition, cyclist, energy, features creation, AdBoost, texture...