Research On Intelligent Monitoring System Of Coal Preparation Plant Based On Vision

The coal security has attracted much attention recently. Coal dressing is anessential process in coal production, where accidents occurring frequently. This thesisadds visual elements to the existing video surveillance system of coal preparationplant. Upgrade it to an intelligent monitoring system which can carry on humanmovement detection, track independently, and alarm when something abnormal, etc.The mian task of this thesis includes:a. Moving target detection. Take image noise processing for video frame. It’sthrough the background subtraction that based on mixture Gaussian model to detectmotion target, then accomplish target segmentation roughly according to theproportion of human body.b. Feature extraction of human image based on6-LBP and HOG. This thesiselaborates and makes use of two feature extraction methods, ones is the HOG featureswhich based on the edge gradient of appearance and outline of local target, and theother one is the LBP feature which based on the color and texture information of localarea. Another feature extraction method,-LBPis proposed on the basis of thetraditional LBP feature extraction methods, remove part of the redundant informationand retain more edge profile information of human body. Finally, use the PCA methodto reduce the feature dimension. Experiments show that the feature extraction methodbased on HOG combining with6-LBP feature has good characterization result for thehuman body.c. The design of human classifier via sparse representation. Considering thesparse representation theory has discrimination which can choose the effective subsetand reject invalid subset, the thesis use sparse representation method that gets thesparse solutions via1-minimization to design the human classifier. Experimentsshow that the human classifiers based on sparse representation has better effect thansupport vector machine (SVM) in low-dimensional feature. Especially when theocclusion happens it achieves a satisfied effect.d. Combined CamShift algorithm with Kalman prediction for moving humantracking. CamShift algorithm is improved based on the Mean Shift algorithm. It hasthe advantages of simple calculation and strong robustness. Through the Kalman filterforecast the human movement parameters, this thesis overcomes track loss that causedby the background interference and occlusion problem. Experiments show that the CamShift tracking algorithm after Kalman prediction is fast and accurate.e. The design and implementation of the intelligent monitoring system of coalpreparation plant is the final work. The system is divided into four modules: systemmanagement module, information processing module, information storage module,and user interaction module. Finally, the system is realized based on C#in the.Netplatform, EmguCV visual class library, and SQL Server database.