A constructing vehicle intrusion detection algorithm based on BOW presentation model

Abstract

Existed algorithms to detect constructing vehicles intrusion in order to prevent power grid transmission line from damage by using video/image processing can only detect constructing vehicles in a single color. In this paper, a new algorithm to detect invasive constructing vehicles based on BOW presentation model is proposed. Firstly, Gaussian fuzzy operation is imposed on the image and Gaussian mixture modeling method is used to separate the foreground regions from background. Then dense SIFT features are extracted from the foreground regions and the features are quantified by using visual dictionary which have been studied previously. Next, the SVM classifier based on histogram intersection kernel function is applied for vehicle type recognition. Finally, duplicated vehicles are removed and alarming signal is sent to workers who need to deal with the hidden damage actions. The experimental results show that the proposed method can effectively detect large constructing vehicles of different colors and categories.