Motion Detection And Its Application In Coal Mine Surveillance Video

As a main kind of motion detection algorithm,background modeling is applied in many areas.To solve the problem of motion detection under sudden illumination change and to reduce time cost,three algorithms are proposed in this paper.Moreover,proposed algorithms are applied to underground people detection and large foreign objects detection in the coal mine underground surveillance video,providing an effective method to ensure the safety of people and equipment.1.Aiming at detecting moving target under sudden illumination change,a background modeling algorithm based on image standard deviations is proposed in this paper.For the standard deviation of pixel intensities in small neighborhoods cannot be greatly changed by illumination changes,standard deviations of pixel intensity in four directions are calculated and modeled.Adaptive parameter update rate is adopted to speed up model convergence.Experiments show that proposed algorithm can detect correctly when illumination changes greatly with much less noise.Underground people can be detected by motion detection algorithms,but usually the detection is very terrible because the lamp on helmet can greatly change illumination.So,proposed algorithm is applied in underground people detection to protect people from dangerous.2.By taking advantage of the characteristic of fixed camera that the scene keeps unchanged,a new convolutional neural network combined with the idea of background modeling is proposed.In this algorithm,a six-channel matrix composed of a background image and a to-be-detected image is brought into the network.By this way,most background pixels can be excluded in the first several convolutional layers by comparing the difference between the background image and the to-be-detected image.So,the features need to be learned by convolutional layers can be reduced greatly,means that the convolutional neural network requires much less training samples and have a better generalization ability.Proposed algorithm is applied in underground people detection.With a training set contains 337 marked image of one same scene,proposed algorithm can detect people in scenes different from training images.Experiments show that proposed algorithm has great generalization ability and requires very few training samples.3.In order to speed up the Gaussian mixture model,A fast Gaussian mixture model based on pixel intensity classification algorithm is proposed in this paper.Proposed algorithm roughly extracts foreground by pixel intensity classification algorithm,then models foreground with Gaussian mixture model.In this way,the computational cost of identifying and updating the parameters of Gaussian mixture model in the irrelevant background region is omitted and the speed of algorithm is greatly increased.Parameter selective periodic reset strategy in some special scenes is introduced to proposed method to maintain detection capability.Experiments show that the proposed algorithm is about 20 times faster than Gaussian mixture model and has much less noise.Moreover,proposed algorithm is applied to the detection of large foreign bodies on belt conveyor to ensure the safe operation of belt conveyor while reducing the calculation cost as much as possible,So that one CPU can process multiple high-definition videos at the same time.The proposed detection method for large foreign bodies has been operating in two coal mines and has achieved good results.