A Real-Time and Robust Distribution Fitting Method for Near-Infrared Surgical Marker Point Localization

To enhance the tracking accuracy and real-time performance of optical tracking systems (OTS) in minimally invasive procedures, we propose a real-time and robust distribution fitting method for near-infrared surgical marker point localization. The method models the marker spot's intensity distribution with an overexposed center and a Gaussian transition at the edges, and employs a subpixel centroid extraction technique based on fast Gaussian fitting along the spot edge. During field-programmable gate array (FPGA) deployment, a single-pass edge detection scheme is implemented to reduce latency and computational overhead, while the positive symmetry of matrices in the Gaussian fitting process is exploited to improve efficiency. Additionally, a dynamic-shift mechanism is introduced to mitigate accuracy loss and reduce resource consumption. Simulation and experimental results demonstrate that the proposed method achieves a tracking frame rate of 95 Hz with significantly higher localization accuracy compared to the conventional gray-weighted centroid(GWC) method commonly used in OTS. The approach maintains robustness under realistic imaging conditions and is well-suited for real-time hardware implementation. This study presents a practical and scalable solution for enhancing optical tracking systems in image-guided surgery, contributing to more accurate and reliable surgical navigation.