Traditional Go stone recognition methods often fail under complex illumination scenarios and are not lightweight enough for an embedded deployment. In order to address this problem, this paper proposes a real-time recognition system for 3×3 Go game boards that integrates dynamic illumination compensation, hybrid edge detection, and multimodal low-dimensional feature fusion. This system employs precomputed lookup tables for Gamma correction, hybrid edge detection based on Sobel gradient fusion and a Canny operator with 3×3 rectangular kernels, and the weighted fusion of wavelet features and compressed LBP descriptors for reducing the computational overhead while maintaining robustness to illumination changes. The experiments carried out on a 900-image dataset across high-intensity illumination, indoor lighting, and low-light conditions achieved an average accuracy of 97% with an inference time of 46 ms, improving the F1 score by 9% in comparison with the conventional methods based on the Otsu algorithm. The ablation studies confirm the contribution of each module of the proposed system, showing an accuracy improvement from 74% to 97% under low-light conditions. The proposed framework balances accuracy and efficiency, demonstrating a strong potential for embedded Go robots and broader lightweight recognition applications (edge computing devices) in dynamic environments.
Go game board recognition, Adaptive threshold segmentation, Hybrid edge detection, Multimodal feature fusion, Lightweight feature extraction.
Yuanshuai LAN, Qian WANG, Xue-qing MENG, Chuan LI, "Lightweight Adaptive Thresholding for Real-Time 3x3 Go Game Board Recognition Under Dynamic Illumination", Studies in Informatics and Control, ISSN 1220-1766, vol. 35(2), pp. 65-75, 2026. https://doi.org/10.24846/v35i2y202606
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).