Computer Vision Based Automatic Recognition of Pointer Instruments: Data Set Optimization and Reading.

Area of Science:

• Robotics and Automation
• Computer Vision
• Industrial Intelligence

Background:

• Intelligent substations increasingly utilize robots, necessitating automated industrial processes.
• Existing meter reading methods struggle with complex environments, hindering accurate data extraction.
• Challenges include reliably detecting meter areas and pointer centerlines for industrial applications.

Purpose of the Study:

• To develop an automated method for reading pointer-type instruments in industrial settings, specifically intelligent substations.
• To overcome limitations of current meter reading techniques in complex background environments.
• To improve the accuracy and efficiency of meter readings in substations.

Main Methods:

• Object detection using Faster Region-based Convolutional Network (Faster-RCNN) to identify instrument panels.
• Data augmentation via Poisson fusion and dataset optimization using K-fold cross-validation.
• Image preprocessing followed by pointer centerline detection using the Hough transform.

Main Results:

• Achieved improved accuracy in target detection through dataset optimization techniques.
• Successfully detected pointer centerlines for accurate meter readings.
• Demonstrated suitability of the proposed method for automatic meter reading in substation environments.

Conclusions:

• The integrated Faster-RCNN and computer vision approach offers a viable solution for automated pointer meter reading.
• The method effectively addresses challenges posed by complex substation environments.
• This research provides a feasible framework for target detection and reading of pointer meters in industrial applications.