Deep Learning-based Recognition Method of Pointer Instrument for Inspection Robot
Aiming at the problems of low precision of instrument identification,low practicability of scene adaptation,and inability of real-time processing of patrol robot,several different methods of me-ter pointer recognition at home and abroad are compared and analyzed.Based on the advantages and dis-advantages of various methods,a method combining an image processing algorithm for region segmen-tation with a deep learning target detection algorithm is proposed to stabilize the automatic recognition algorithm for pointer instruments and enhance its adaptability to the environment.YOLOv5 target de-tection algorithm is used for dial positioning,image smoothing,color correction,and color enhance-ment processing are used to reduce the impact of on-site light on target area positioning.Then,the im-age processing algorithm for region segmentation is used to segment the pointer area and tick mark area,obtaining the maximum and minimum angles of the pointer's rectangular rotation center and tick mark.The instrument reading is calculated based on the pointer angle and range,the pointer extraction capa-bility is improved through image pre-processing and HSV color space transformation algorithms.Based on real-scene testing,the problem of pointer instruments being affected by occlusion,blurriness,and red line warning bars,which makes it difficult to accurately locate the dial and extract the pointer,has been resolved.The experimental results show that this method is not only high in detection accuracy and speed,but also practical,and meets the requirements of robot patrol inspection.
万方数据
维普
