Accurate localization and assessment of burn injuries are vital for informed clinical decision-making and timely treatment, especially in primary care settings with limited specialist access. Traditional visual inspection methods are often subjective and suffer from significant inter-clinician variability. This study proposes a real-time stereo vision framework for precise burn region localization using a binocular camera system and the Semi-Global Block Matching (SGBM) algorithm. The system incorporates Zhang’s checkerboard calibration to determine intrinsic and extrinsic camera parameters, and employs a multi-cue cost computation strategy that combines Sobel edge gradients with Birchfield-Tomasi intensity metrics to enhance disparity accuracy. Post-processing is performed using Weighted Least Squares (WLS) filtering to suppress noise and improve depth map quality. The resulting disparity is reprojected into 3D space to generate a colorized point cloud, allowing intuitive spatial visualization of affected skin surfaces. Experimental evaluation on embedded hardware (Jetson Nano with IMX219- 83 stereo cameras) demonstrates that the system achieves sub-millimeter depth accuracy (<1 mm) in the optimal range of 450–500 mm and maintains an average error below 2 mm across 450–800 mm. Real-time performance is achieved with average per-frame processing times of 0.42 s. These results underscore the system’s potential as a portable, cost-effective solution for high-precision burn localization and continuous treatment monitoring in both acute and rehabilitation settings.