A novel video detection design based on modified adaboost algorithm and HSV model

Abstract

In modern traffic systems, accurate video detection is a key challenge for traffic management. Aiming at the problem of public bus detection, this paper proposes a video detection method to well recognize the buses. Firstly, we employ the foreground detection method to find the moving vehicles. And then a training classifier which consists of the improved Adaboost algorithm and Haar-like features is proposed to filter undesired vehicles. Secondly, we use the Canny operator to locate bus characteristics, and further detect the bus with the modified HSV model. This design is tested on the Visual Stadio and OpenCV platform in which load the urban transport data as the samples. The test results show that our detection method has better robustness than both three-frame differential method and hybrid Gaussian method, and the accuracy of detection on the window positioning is more than 93 percent.