Automatic Visual Reading of Meters Using Deep Learning

Abstract

In this paper, we present a novel approach to the problem of reading residential meters using deep learning algorithms. As a starting point we use Faster R-CNN method and, to acquire more precise readings, we modify its functionality. As there were no databases for this kind of task, one had to be collected and properly annotated. This paper also provides a brief introduction to methods for image augmentation and a technique to augment annotated image dataset. For each part of the presented method as well as the whole method as one unit experiments were conducted to show the overall successfulness.

Towards Image-Based Automatic Meter Reading in Unconstrained Scenarios: A Robust and Efficient Approach

“…Object detectors have been explored frequently to deal with counter detection. For example, Koščević & Subašić [10] employed Faster R-CNN [24] to detect counters and serial numbers on images of residential meters, whereas Tsai et al [12] applied and fine-tuned Single Shot MultiBox Detector (SSD) [25] for counter detection in electricity me-ters. In both studies, only proprietary datasets were used to evaluate the detectors.…”

“…It is important to point out that in several recent works in the literature [10], [15], [21], [30] the authors reported only the results obtained by the proposed methods or compared them exclusively with traditional approaches, overlooking deep learning-based approaches designed for AMR. This makes it difficult to make a fair comparison between recently published works.…”

“…In this context, image-based techniques for AMR are much needed, especially taking into account that it is not feasible to quickly replace old meters with smart ones [10]- [12]. The idea behind image-based AMR, which is an specific scenario of scene text detection and recognition, is that the aforementioned inspection can be carried out automatically, reducing mistakes introduced by the human factor and saving manpower [3], [13].…”

“…Although AMR (hereinafter AMR refers to image-based AMR) has received great attention in recent years, most works in the literature are still limited in several ways. In general, the experiments were performed either on propri-etary datasets [4], [10], [15] or on datasets containing images captured on well-controlled environments [3], [11], [16]. This is in stark contrast to related research areas, such as automatic license plate recognition, where in recent years the research focus shifted to unconstrained scenarios (with challenging factors such as blur, various lighting conditions, scale variations, in-plane and out-of-plane rotations, occlusions, etc.)…”

Towards Image-based Automatic Meter Reading in Unconstrained Scenarios: A Robust and Efficient Approach

“…In this context, image-based techniques for AMR are much needed, especially taking into account that it is not feasible to quickly replace old meters with smart ones (Koščević & Subašić, 2018;Li et al, 2019;Tsai et al, 2019). The idea behind image-based AMR is that the aforementioned inspection can be carried out automatically, reducing mistakes introduced by the human factor and saving manpower (Laroca et al, 2019a;Zuo et al, 2020).…”