Application Of Improved Optical Flow Histograms In The Detection Of The Working Condition Of Hydraulic Guards

Hydraulic support guard plate is one of the important devices of modern coal mine operation.Its main function is to prevent coal seam spalling,ensure the safety of coal mine production and the personal safety of related staff.At the same time,in order to further promote the unmanned operation of the mine and develop the intelligent mine,this paper designs a set of intelligent auxiliary detection scheme for the working state of the protection board based on the motion characteristics of the protection board in the underground monitoring video of the coal mine,and on this basis,a new motion feature extraction method is proposed.The main research work of this paper is as follows :In view of the poor quality of video image in coal mine scene,image pre-processing was carried out on video images: mean filtering and dark channel defogging,and image comparison was carried out before and after image pre-p rocessing to achieve image denoising and defogging to restore images;In order to realize the real-time detection of the working state of the hydraulic support guard plate in the monitoring video,a set of real-time detection scheme for the working state of the guard plate is proposed,which divides the detection of the working status of the guard plate into two modules: the guard plate positioning algorithm and the guard plate working status recognition algorithm.Firstly,a neural network is used to locate the position of the guard plate,and then the motion analysis is carried out in the area of the guard plate,and finally the real-time detection of the working state of the guard plate is realized;For the motion characteristics of the working state of the hydraulic support guard plate,the background modeling and motion feature extraction of video image are attempted.Three motion feature operators are compared:Historgram of Oriented Gradient（HOG）,Motion Boundary Historogram（MHB）and Historgram Oriented Optical Flow（HOF）,O and an improved optical flow histogram descriptor is proposed on this basis.Through the comparison experiments of the detection effects of four descriptors and the comparison experiments of the feature dimensions of different improved optical flow histogram descriptors,the experiments show that the improved optical flow histogram descriptors proposed in this paper are more robust to the extraction of motion features,with the accuracy of the data set reaching 87.77 %,the recall rate reaching 78.19 %,the F1＿score value reaching 75.55 %,and the fps value reaching 18 frames/s.The recognition rates of the three working states（static,upward and downward）of the guard plate are 96 %,84 % and 85 %,respectively,which can basically realize the real-time detection of the working state of the hydraulic support guard plate in the coal mine scene.