Research On Human Detection And Tracking In Underground Coal Mine Videos

There are some dangerous regions in the underground coal mine, the miners are not allowed into these regions. Considering the requirements for the safety production of coal mine, the methods of human detection and tracking in underground coal mine videos are studied in this paper. The main research work includes:A fuzzy enhancement method was proposed to overcome the impact of low-illumination and uneven lighting. A linear fuzzification function was constructed; and the image was enhanced fuzzily to enhance dark regions and to restrain the glaring regions; then, the contrast of the image enhanced was adjusted. The experimental results show that the method produces better result.A fuzzy miner detection method was proposed. The difference image and the frame image were fuzzificated. Some fuzzy rulers according to the characters of coal mine videos were defined to get every pixel's membership value in the object. The silhouette and its membership function were defined based on human vision. Two functions were combined to detect the miner object. On the other hand, the fuzzy detection was combined with the mixture Gaussian model to detect the miners in dynamic scene. The experimental results show that this method can remove the interference of miner's lamp and detect the miner effectively even they are similar to the background. This method has the characteristics of less calculation and high velocity; it is suited for practical use in underground coal mines.A method for detecting miners based on helmets detection was proposed. If a helmet is detected, it means that a miner is detected. The method constructed the standard images of helmets, extracted the four directional features, modeled the distribution of features using Gaussian function, designed piecewise linear classification, and separated the local image of frames into helmet and non-helmet. The experimental results show that this method can detect the helmets effectively.Helmet tracking method was proposed. The tracker adopted Kalman-Meanshift scheme, constructed joint histogram based on edge orientation and position information, selected Kernel-bandwidth automaticly according to the oval shape of helmet, and tracked helmet accurately and real-timely.