Research On Safety Monitoring System For Workers Of Train Maintenance Garage Based On Machine Vision

Since there are many kinds of workers and different range of dangerous areas in trainmaintenance garage, it is particularly important to ensure the safety of workers. The existingvideo monitoring system of train maintenance garage is the important way to ensure that theworkers are safe. Machine vision technologies bring the unprecedented development to allkinds of video surveillance systems, it is of important practical significance that the machinevision technologies are applied to safety monitoring for workers in train maintenance garage.Moving target detection and recognition are the key technologies of machine vision. Forthe special application scenarios of train maintenance garage, the research focus of this thesisis that the two key technologies are studied on theoretical analysis and experimental research,the software of safety monitoring system for workers of train maintenance garage based onmachine vision is designed and achieved on the basis.The existing various kinds of moving target detection algorithms are researched andcompared, given that there is complex condition of light in live train maintenance garage, adetection algorithm that combined Gaussian Mixture Model (GMM) with texture feature isadopted. At the beginning of the target extraction, the idea of light detection is joined, whenthere are no illumination changes or small illumination changes in images, the target detectionis implemented through the method of GMM algorithm, when there are fast illuminationchanges in images, the target detection is implemented through the method of texture featurealgorithm, thus the moving target detection is carried out through combining the advantagesof the two algorithms. The experimental analysis shows that the influence of illuminationfactor is basically eliminated by this method, the good effect can be obtained especially in thecondition of fast illumination changes.On the basis of analysing the existing moving target recognition algorithms, SVM(Support Vector Machine) is used to recognize workers of train maintenance garage for thescenarios of train maintenance garage. Different kinds of classification features of human areanalyzed, HOG (Histograms of Oriented Gradients) of the moving target binarization imagesand shape parameters of the moving target areas are selected as the identification features ofSVM.Human recognition is more efficient by HOG and shape features of moving targets, andthat the target recognition effect that is caused by some error detection can be avoided. Theclassification features of positive and negative sample images are entered into SVM, thereal-time acquisition images collected by the system are recognized by off-line trainingclassifier.Experimental results show that the moving target recognition is achievedeffecttively by this algorithm.The software of safety monitoring system for workers of train maintenance garage based on machine vision is designed and achieved with functions of video capturing, dangerous areasetting, moving target detecting, moving target recognizing, etc. The detection andrecognition can be carried out for workers who inadvertently stray into a given dangerous area,the alarm is executed if the result of recognition is a person, the functional requirements ofthis system are completed.