Research On Dynamic Distance Estimation Of Substation Foreign Object Intrusion Based On Binocular Stereo Vision And Deep Learning

Foreign object intrusion detection in substations mainly relies on manual inspection and traditional image processing methods,which have problems such as low efficiency,poor accuracy and poor real-time performance.With the development of unattended substation,how to effectively detect and prevent foreign body intrusion has become an important research topic.The ranging technology based on machine vision has the advantages of non-contact,real-time,flexibility and accuracy.It can automatically determine whether there is foreign object intrusion in the substation through image analysis and recognition.Based on this,this paper proposes a dynamic distance estimation method for substation foreign object intrusion based on binocular stereo vision and deep learning.Firstly,aiming at the problems of distortion between cameras and noise and impurities in the image,histogram equalization,gamma correction and image filtering are performed on the image to enhance the robustness of the matching algorithm.Secondly,through camera calibration and epipolar constraint correction,the internal and external parameters of the camera are obtained,which simplifies the process of matching and depth calculation.Aiming at the problems of low texture area,occlusion area and illumination change in the picture,it is easy to have mismatch and low accuracy.In the stereo matching stage,the BM algorithm,SGBM algorithm,SGBM algorithm based on Census transform and depth map based on deep learning monodepth algorithm are analyzed and compared.From the results,the monodepth algorithm based on deep learning performs best and realizes real-time stereo matching under different scenes and illumination conditions.In view of the complex and changeable environment inside the substation and the unknown characteristics of foreign bodies,it is easy to cause false detection or missed detection.By constructing the foreign body data set of the substation,the YOLOv7 algorithm is used to identify the foreign bodies in the substation.The results show that the average mAP of the model for nine categories reaches 87.2%,and the real-time detection time of each image is only 0.0134 s.The model evaluation index is compared with the classic one-stage algorithm SSD and the two-stage algorithm Faster R-CNN to further verify its superiority.In order to verify the feasibility of the proposed method,the field simulation experiment of target recognition and dynamic ranging is carried out in the substation.The results show that the proposed method can realize real-time dynamic ranging with relative error less than 10%within 5 meters on the basis of accurate target detection.The method proposed in this paper can accurately identify foreign objects in the substation environment and estimate the distance in real time,thereby reducing manual intervention and maintenance costs and improving the intelligent level of the substation.