Equipment Status Recognition And Instrument Reading Based On Monocular Stereo Visio

In gas power stations,the application and recycle of gas are likely to cause accidents such as explosions due to the characteristics of the gas.In order to ensure gas power stations can operate safely and efficiently,it is significant to identify the status of the equipment in the gas power station and detect the readings of the instruments.This thesis proposes a monocular stereo vision-based equipment condition detection method to monitor equipment in gas power stations from the following four steps.Firstly,the equipment types are identified through the target detection network;secondly,different operations are performed on the different types of identified equipment,and the identified electric generator is detected to judge if it is on fire or smoke;thirdly,the pointer instrument is photographed obliquely by the industrial camera with the designed four-mirror stereo adapter to obtain two horizontal parallax images which are used to reconstruct the pointer instrument,and the angle method is applied to obtain readings so as to reduce the reading error caused by tilting;fourthly,another three steps are applied to obtain readings of the digital display instruments:firstly,the images of digital display instruments photographed obliquely are corrected through the adoption of perspective transformation;secondly,the characters are split through the application of connected domain;finally,the characters are identified through the adoption of the threading method,and the identified results are connected in order and output to obtain the readings.The thesis consists of the following work:(1)Compare single-stage and two-stage target detection algorithms,and select the YOLOv5 model as the model for the identification of device type and state.(2)Construct the database of the images of equipment,flame and smoke,and the data set is expanded through data enhancement methods such as rotation and brightness adjustment.(3)Train the YOLOv5 network model to identify equipment,flames and smoke through external surveillance cameras.(4)Design a four-mirror stereo adapter,complete its model simulation and physical construction,and use the four-mirror stereo adapter and a monocular camera to obtain two horizontal parallax images to achieve three-dimensional reconstruction and to obtain the readings of pointer instruments.(5)The images of digital display instruments photographed obliquely are corrected through the adoption of perspective transformation,and suitable character segmentation and character identification methods are used to read the instrument.