Chinese thesis title: Research and Application of Pointer Meter Reading Recognition Algorithm Based on Key Point Detection
English dissertation title:
Research and Application of PointerMeter Reading Recognition AlgorithmBased on Key Point Detection
Partial summary :
On the basis of summarizing the key point detection and automatic reading recognition methods of traditional pointer instruments, this paper designs a key point detection algorithm based on deep learning and applies it to the automatic reading recognition task of pointer meters. The main research achievements of this paper are as follows:
1) A key point detection algorithm based on MaskRCNN is proposed. By adding a deviation branch on the basis of the MaskRCNN network structure and correcting the key point position deviation caused by the upsampling of the original network, the positioning accuracy of the network in the key point detection task is improved. The experimental results on the public data set show that the algorithm proposed in this paper has improved in all indicators, effectively improving the positioning accuracy of key points.2) An algorithm for automatic reading recognition of pointer instruments based on key point detection is proposed. Combining the traditional automatic reading recognition task of pointer instruments with the key point detection task based on deep learning, an algorithm framework for automatic reading recognition of pointer meters based on accurate key point detection is designed, and the key point detection algorithm is used to locate the scale line and pointer position, thereby Further realization of reading recognition. The experimental results show that the algorithm proposed in this paper has high accuracy in the recognition of various pointer instruments, and has good robustness to shooting angles and lighting conditions.
1. Introduction
This chapter takes various types of pointer instruments in substations as the research object, and explores the method of automatically identifying instrument readings. The key to the automatic reading task of the pointer instrument lies in the positioning of the pointer instrument and the angular relationship between the pointer and the scale point.
2. Testing process
1. Use the improved Mask RCNN key point detection algorithm proposed in Chapter 3 to accurately locate the scale points of the pointer instrument and the feature points on the pointer.
Then use the characteristic points of the pointer to fit the straight line where the pointer is located, so as to use the angle relationship to judge the number of the instrument.
2. Extract key points
The idea of key points, when establishing the pointer instrument data set, this paper regards the dial as a whole, and each scale point on the dial is marked as a key point, and the visual result of the mark is shown in Figure 4-5 below. For each scale line of large range in the meter, take the endpoint of the scale line as a key point. For the straight line of the pointer, the three positions of the head of the pointer, the center of gyration and the tail of the pointer are taken as key points, which are used to locate the position of the straight line of the pointer. Take Figure 4-5 as an example. The gauge is a pressure gauge in a substation with a range of 0 to 1. There is a large scale line every 0.1 range. Take the end of the scale line as a key point to indicate the scale line There are 11 key points in total, expressed as sl, s 2,..., s 11; for the pointer, take its top as the first key point p1, the center of rotation of the pointer as the second key point p 2, and the pointer The middle position of the tail serves as the third key point p3. Therefore, for the table in Figure 4-5, there are 14 key points in total, including 11 tick marks and 3 pointers. For other types of meters, use the same principles to select critical points.
For the pointer instrument key point data set marked as above, after the key point detection network is input to complete the model training, the instrument dial will be detected for each input pointer instrument picture during the test and all the key point information in the dial will be obtained . Afterwards, based on these key point information, the automatic reading of the pointer instrument can be completed.
3. Then the author uses the extracted key points to perform arc fitting.
5. The pointer is positioned in a straight line, and the scale is calculated using the angle.
The angle method identifies the readings of pointer instruments by measuring angles. While extracting the instrument pointer in the previous section, the intercept and slope of the straight line where the pointer is located are obtained, and the direction of the pointer is also determined. For each instrument, first calculate the angle relationship between the instrument pointer and the zero scale line through the slope, and then convert the instrument representation number through the instrument's range and angle relationship. Suppose the range is M, the angle between the pointer and the zero scale line is a, and the angle between the zero scale line and the full scale is B, then according to the angle method, the meter reading V is:
6. Experimental results
3. Summary
This paper proposes an improved accurate key point detection algorithm based on Mask RCNN. Compared with the original Mask RCNN, the network designed in this paper has higher accuracy in key point detection, and accordingly proposes a pointer instrument based on key point detection. Automatic reading recognition algorithm framework, which has the characteristics of high accuracy, good robustness, and wide applicability. The work done in this paper is summarized as follows:
1) This paper introduces the research background and significance of automatic reading recognition of pointer meters, summarizes the research status and content of this field and the field of key point detection at home and abroad, and summarizes the automatic reading recognition of pointer meters. The four steps required by the algorithm are identified, and the existing methods are elaborated step by step.
2) This paper reviews the development history of human key point detection or pose estimation, focusing on the network structure and loss function of Mask RCNN, a classic algorithm in the field of key point detection. The improvement direction of point detection accuracy, a more accurate key point detection algorithm based on Mask RCNN improvement, has achieved good results on public data sets, and a simple analysis of the impact of multi-task learning on network performance .
3) This paper combines the traditional automatic reading task of pointer instruments with the key point detection task based on deep learning, and proposes an automatic reading recognition algorithm framework for pointer instruments based on accurate key point detection. Automatic reading recognition. Experiments prove that this algorithm has greatly improved in accuracy, robustness and practicability compared with traditional algorithms.
The algorithm proposed in this paper is suitable for various types of pointer instruments in substations, and there are no restrictions on the position, angle and ambient light of the pointer meters in the field of view when shooting, and can be captured by smart devices such as inspection robots or drones. , the image acquisition cost is low. Compared with the traditional pointer instrument recognition algorithm, the method in this paper has stronger robustness and applicability, and the accuracy meets the accuracy requirements of the subject system design, and can be applied to the real operating environment of the substation.