Safety Helmet Detection Algorithm based on Color and HOG Features

Abstract

There are many unstable factors in the working environment of electrical workers, which threaten their safety. Therefore, how to protect electrical workers is a problem worth thinking about. Safety helmet protects worker’s head from injury when they fall. This paper presents a method to detect whether electrical workers wear a safety helmet or not. This method is based on Support Vector Machine (SVM), the grids of Histograms of Oriented Gradient (HOG) features, and color features. Firstly, we get information about the worker’s work scene image. According to the acquired image, we use the Deformable Parts Model (DPM) algorithm to extract the worker’s area. In the area where the worker exists, we use the method of color space conversion and color feature matching to extract the area where the safety helmet may exist. In this area, we use the SVM trained by HOG features to detect the safety helmet and ultimately to realize the judgment of workers wearing a safety helmet. In the experimental part, the effectiveness of our method is demonstrated. Compared with the Color + CHT and Color + Number of Pixels, our method has been improved by 3 to 4 percentage points.