Research On Empty Subway Detection Based On SVM

As the principal mode of transportation in our country, the operation mode of urban rail transit system has been changed, with the progress of science and technology, and the rapidly increase of the degree of automation. Subway trains transit to fully automatic operation(FAO)from the traditional manual driving, higher requirements on safety, reliability and service level of the subway train are put forward.In traditional, detection of abandoned objects in driverless trains is done by the staff when the train stops for arrivals to station, which makes it not only time-consuming, but also is heavy labor intensity and overloaded work. As the intelligent video, image processing and pattern recognition technology have been constantly improved and perfected, it provides the reliable theoretical basis for empty subway detection of FAO trains by machine.By analyzing the integrated supervisory control system(ISCS) composition and function of FAO trains, empty subway detection work is automatically done by using inner videos collected with ISCS, to realize image pretreatment, background modeling, image feature extraction and judgment for empty subway detection and so on. The main research content is as follows:Firstly, in the image preprocessing stage, considering the influence on obtained image caused by environmental conditions and the collection and transmission equipment, an improved salt and pepper noise filtering algorithm is proposed for the noise elimination, and an adaptive image enhancement algorithm based on matching pyramid decomposition is proposed for image enhancement. The analysis and comparison of the algorithms performance are made from subjective visual effect and objective evaluation criteria.Secondly, in the image feature extraction process, advantages and disadvantages of the various features are analyzed, and the human heads are choosed as the features extraction part,and histogram of oriented gradients(HOG) of head, and hue, saturating, value(HSV) color space feature of facial characteristics are used for feature extraction, to process the characteristics after fusion by dimension reduction using principal component analysis(PCA).Finally, in the empty subway detection phase, the first is background modeling using the averaging method. The foreground is extracted by using the background subtraction method,using support vector machine(SVM), adaboost and neural network classifier to classify the local binary pattern(LBP) features, HOG features and fusion HOG+HSV features processed by PCA dimension reduction, respectively. Finally, SVM classifier is adopted to recognize the empty subway detection results.