Video Monitoring Technology Under The High Voltage Transmission System

The research object of this paper is pedestrian detection and recognition in high voltage transmission system.The monitoring device of high voltage transmission system is generally installed on the high voltage transmission tower more than 60 meters away.In the picture taken by the video monitoring probe,there are too few pedestrian pixels,so it is impossible to use the traditional pedestrian detection algorithm to identify whether there are pedestrians in the image frame.This paper proposes a targeted,accurate and efficient solution to this problem.Firstly,the difficulty of pedestrian detection and recognition in the background of high-voltage transmission is that there are too few pedestrian pixels,which leads to the low recognition accuracy of traditional detection algorithms.The difference area can not be obtained by comparing the front and back frames,because the environment of the camera is easily affected by natural factors such as wind and ground vibration,resulting in slight jitter in the camera picture.In view of this difficulty,this paper proposes a new image registration algorithm based on SIFT algorithm.The algorithm mainly calculates the angle difference and offset difference of two images by SIFT algorithm.After the corresponding geometric operation of the image,the two images can be overlapped and put forward to the difference area.Experimental results show that,compared with the traditional image registration algorithm,the improved video stream difference region detection algorithm based on SIFT algorithm has higher recognition accuracy.After obtaining the difference area,this paper proposes a set of mathematical model to calculate the adjustment parameters of PTZ.According to the relative position of the center of the difference region in the captured image,the rotation angle of the pan tilt are calculated,and the pan tilt is operated according to the parameters.Then,compared with before focusing,the pixel number of pedestrians in the zoomed image increases a lot,but the pedestrian features in the top view angle are much less.In view of the target detection task under the top view angle,this paper compares a variety of detection models based on hog feature extraction algorithm.Firstly,the classical pedestrian detection algorithm hog + SVM model is reproduced to show the detection performance of the model in the dataset.Then the training model is improved,and the SVM training model is replaced by Ada Boost model,namely HOG+Ada Boost model,which also shows the detection performance of the detection model on the test image set.Finally,a method of replacing the training model with the more powerful XGBoost model,namely HOG+XGBoost,is proposed,and the optimal super parameters of the model are found through the grid search method,and the detection performance of the model on the test set of this paper is demonstrated.The experimental results show that,compared to the support vector machine and the adaptive enhancement algorithm in ensemble learning,the HOG+XGBoost detection model after replacement and tuning shows high-precision detection performance on the data set of this paper.