.png){kind=icon}
Return
Structured Visual Measurement and Consistency Correction for Pointer Scale Systems
.png){kind=icon}
PRE
AI
DOI：10.1109/tim.2026.3723445.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Pointer scale systems are widely used in industrial inspection, power equipment monitoring, and laboratory instrumentation. This article focuses on pointer scale systems with a single pointer and a single linear scale, a common and representative class in practice. For such systems, recovering a quantitative measurement value constrained by the underlying scale structure from image observations is a key task for reliable automatic reading. Existing pipeline-based methods are susceptible to error accumulation, while direct numerical generation approaches lack explicit measurement semantics and consistency constraints. To address these limitations, we reformulate the task as structured visual measurement. We define an intermediate structural representation that describes the measurement configuration of a pointer scale system and build a two-stage structured visual measurement framework (SVMF) on top of it. Stage 1 performs structured visual parsing. Stage 2 corrects structural inconsistencies with the aid of scale value evidence. A deterministic recovery function then produces the final reading. To improve generalization to variations in measurement structure, we further propose a synthetic data construction strategy conditioned on structure and driven by coverage of the measurement space. Experiments show that the default SVMF clearly outperforms the compared open-source baselines on out-of-distribution real data, with an MNAE of 0.0668 and an Acc@0.05 of 74.88 %. The method also exhibits milder degradation under controlled structural shifts and stable behavior under real illumination and pointer position perturbations. Moreover, the explicit structural representation allows structural inconsistencies and potential failure cases in the reading process to be further checked and diagnosed, thereby providing an empirical basis for reliability and risk analysis of the proposed method.
Keywords:
Consistency correction
measurement space
structured visual parsing
vision-based measurement
vision–language models (VLMs)
Journal
IF:
5.9
Papers:
2.0W
Citations:
5.8W
Organization
Cited Papers
No cited papers available