Research On Dection For Human Invasion Of Traction Substation Based On SVM

The safe operation of traction substation, which is an important part of traction power supply system, is significantly important for the stability of the whole power supply system. In order to ensure the safe operation of some important apparatuses in substation and avoid people and animals approaching, which will probably interfere with the operation environment, detecting the invading human is of great importance through image detecting and recognition method.There are lots of methods to detect and recognise human target, among which machine learning and pattern recognition are hot subjects. With this method, invading human targets detecting and recognition are studied on MATLAB platfom using LIBSVM and SVM tool boxes in this paper.Image pre-processing, extraction of image gradient histogram characteristics and the SVM training, testing and prediction are the main content of invading human detection process. Gamma correction, gray scale and filtering processing are used for the detection pictures gained in traction substation, which lays a foundation for verifying the correctness of characteristics extraction and recognition in the following chapters. What is more, the principles of how HOG characteristics describing target invading human during the extraction process are studied and singular data are classified, which provides a theoretical support for following research. In the process of SVM training, test and prediction, study of the training and classification principle and algorithm of SVM is furthered and an effective invading human detection system is established, besides the optimal parameter configuration is chosen through the comparison of different classification effects of SVM under different parameters by means of some necessary experiment tools, which is also verified through the classification results of testing set. Finally, invading human detecting experiment is realized according to the methods and steps designed.In view of the traction substation special scenario, differential area classification and recognition method is researched, and the image within the differential area are obtained through differential operation followed by the extraction of HOG characteristics of this image. Classification experiment on the SVM classifier trained is conducted, which verifies the effectiveness of the method used. And a GUI interface is designed to realize the detection method in the end of the paper. Finally, detection and location for human target in sequential images from the recorded videos are studied, at the same time denoising and shadow elimination are considered, detection and extraction for human target from a episode of video are realized by means of motion detection.