In order to reduce the workload of staff, break through the data isolation between the front and back ends and enhance the foundation of intelligent construction, Research is carried out on pointer-type pressure gauge which is installed on the gas distribution cabinet based on machine vision to realize the automatic value identification of pointer-type pressure gauge. Using YOLOv3 deep learning framework, the recognition model of multi-pressure gauges under complex conditions is trained, and the automatic extraction algorithm of key information of pointer-type pressure gauges is designed by using the universal structural characteristics of pointer-type pressure gauge. The algorithm we proposed is used to extract and recognize the pointer-type pressure gauge on image and give a reading result automatically. We evaluate the model and calculate the accuracy in a video of pointer-type pressure gauge with changing reading. The calculation and analysis results show that the accuracy of the pressure gauge recognition model established in this paper is 100 % in 66 test pictures, and there is no error detection and missed detection. The dynamic video analysis on pressure gauge with range of 2.5MPa shows that the maximum error of the readings is 0.052 MPa, and the average error is less than 0.01 MPa. Our algorithm is implemented in Python language, which is convenient for the deployment of domestic equipment and has certain popularization and application value in future.