To address the challenges of instability and poor robustness in stereo vision beacon recognition under low-texture and high-contrast conditions, this paper proposes a binocular vision positioning system that employs square LED array beacons. A lightweight beacon recognition algorithm is developed, integrating structural perception and stereoscopic matching. The LED array structure is perceived through connected component analysis, polar angle sorting, and structural consistency verification. A joint cost function incorporating brightness difference, disparity consistency, and structural morphology consistency is then constructed to perform stereoscopic matching. Finally, three-dimensional position information is recovered via coordinate system mapping. Experimental results indicate that the proposed algorithm achieves a recognition accuracy of 98% with a reconstruction error of less than 4 cm. Compared to conventional methods such as SGBM and LoFTR, it exhibits superior performance in challenging scenarios involving brightness interference and overlapping targets, while maintaining real-time operation and suitability for embedded deployment.