A REAL-TIME DETECTION METHOD FOR PAIRED ILLUMINATED MARKERS BASED ON COLOR CHANNEL SUBTRACTION AND GEOMETRIC CONSTRAINTS

ABSTRACT

Real-time detection of paired luminous markers is important for embedded vision systems. To address the limited computing resources of embedded platforms, a lightweight detection framework is proposed. The method employs color-channel subtraction for target segmentation, geometric features for light-bar filtering, and predefined constraints for target pairing. Compared with an HSV-based approach, the proposed method achieves higher detection accuracy and better real-time performance. It provides reliable localization of luminous markers with low computational cost and supports subsequent target recognition, tracking, and control tasks.

Keywords: Luminous Marker Detection; Embedded Vision; Color Channel Subtraction.

At present, methods for detecting paired luminous markers can generally be divided into two categories. The first category is based on deep learning, such as using models like YOLO to achieve highly accurate and adaptive target detection[1, p. 57]. However, these methods usually require a large number of model parameters and substantial computational resources, making it difficult to guarantee real-time performance on resource-constrained embedded platforms. The second category relies on traditional computer vision techniques, in which target detection is performed by extracting and analyzing representative features of the target. Compared with deep-learning-based approaches, these methods significantly reduce computational costs and are able to achieve a balance between real-time performance and detection accuracy on embedded platforms.

In traditional computer vision, color segmentation is a key step that determines both detection speed and robustness[2, c. 23]. The most common approach is to transform the image into the HSV color space and extract target colors based on hue thresholds. However, this method often becomes unstable when dealing with overexposed self-luminous objects. Therefore, under the same cascaded pipeline framework, this paper adopts a method based on color-channel subtraction and geometric constraints, and compares it with the HSV-based color segmentation approach. The comparison is conducted to evaluate the performance of the proposed method in terms of real-time capability and detection accuracy.

Table 1.

Performance comparison of different detection methods

As shown in Table 1, the HSV-based method achieves a frame rate of 29 FPS, while the color-channel subtraction method achieves 18 FPS. Both methods satisfy the real-time requirements.

The HSV-based method achieves a 100% recall but only 66.67% precision, meaning that it detects all true markers but produces a considerable number of false detections. In contrast, the color-channel subtraction method achieves 100% precision with a 77.78% recall, indicating that it produces no false detections but misses some true markers. In terms of the overall F1-score, the color-channel subtraction method reaches 87.5%, outperforming the HSV-based method and demonstrating a better balance between precision and recall.

This paper implements a lightweight detection framework for paired luminous markers based on color-channel subtraction and geometric constraints, and compares it with an HSV-based detection method under the same processing pipeline. Experimental results show that the color-channel subtraction method provides more reliable detection results and significantly reduces false positives, making it particularly suitable for embedded vision systems with limited computational resources and high costs associated with false detections. The main limitation of the proposed method is the reduced recall caused by missed detections when the central regions of some markers become overexposed. Future work will introduce adaptive chromatic-difference thresholding and overexposed-core recovery strategies to further improve the recall rate.

References:

1. JIKE XU. Research on Machine Vision-Based Target Recognition and Localization Methods[D]. Xi'an: Xi'an University of Science and Technology, 2019.
2. YANGJIAN GAO. Target Localization in Mobile Robots Using Omnidirectional Vision and Combined Color–Shape Features[D]. Chongqing: Chongqing University, 2017.