Lightweight machine-vision framework pinpoints EV charging holes without GPU training

Researchers have developed a lightweight machine-vision framework that locates the charging holes on an electric vehicle’s charging interface without any learning-based training process, reporting an overall mean localization error of 4.2 pixels and a 96.25% detection success rate on conventional CPU hardware. Accurate localization of these holes is described as an indispensable perception step for automatic charging systems, because the success of subsequent charging-gun insertion depends directly on precise estimation of the geometric centers of the holes on the vehicle charging interface.

The framework, presented by Wei Luo, Qiantao Huo, and Xin Li, integrates image filtering, RGB-to-HSL color-space conversion, parameterized HSL adjustment, two-stage contour detection, minimum enclosing circle fitting, and Hough circle verification into a single deterministic pipeline. Because the method is rule-based rather than learned, the authors describe it as interpretable and deployable on conventional CPU-based computing resources, avoiding the training data requirements and opaque decision-making associated with neural-network approaches.

The image data were acquired by a monocular RGB camera mounted on the charging robot end-effector, the same sensing configuration in which the system is intended to operate. In total, 80 images were collected from six vehicles equipped with the same five-hole charging interface, capturing realistic variations in illumination, viewing angle, reflective interference, and background complexity.

Performance was measured on an Intel Core i7-12700 CPU with 16 GB of RAM and no GPU acceleration. Under these conditions, the framework achieved the reported 4.2-pixel mean localization error and 96.25% detection success rate, with an average per-image processing time of approximately 60 milliseconds. The authors also report detailed parameter selection rationale, a parameter perturbation analysis, and a condition-specific ablation analysis examining how the pipeline behaves across the tested imaging conditions.

According to the study, the results demonstrate that the proposed framework provides accurate, robust, and engineering-deployable charging-hole localization for the target five-hole charging interface. The combination of sub-100-millisecond processing on commodity CPU hardware and a fully interpretable pipeline positions the approach as a practical perception component for automatic electric vehicle charging robots operating in the lighting and reflection conditions tested in the study.

Subject of Research: Machine-vision localization of electric vehicle charging holes for automatic charging systems

Article Title: A machine-vision framework for electric vehicle charging hole localization based on contour-driven detection and geometric verification

Article References: Luo, W., Huo, Q., & Li, X. (2026). A machine-vision framework for electric vehicle charging hole localization based on contour-driven detection and geometric verification. PLOS One, 21(10), e0359570. https://doi.org/10.1371/journal.pone.0359570

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0359570

Keywords: electric vehicles, machine vision, charging hole localization, contour detection, Hough circle transform, HSL color space, charging robots, computer vision, geometric verification, CPU deployment, autonomous charging, machine-vision