Real-Time Feature Detection for Booster Recovery: An Adaptive Harris Framework Integrating Scharr Vectorized Gradients and Dilation-Based NMS

Key words:  Adaptive Robust Harris (ARH)  Scharr operator  adaptive thresholding  non-maximum suppression (NMS)  complex environments  rocket booster recovery

Abstract:

—This paper presents an Adaptive Robust Harris (ARH) corner detection algorithm to address challenges in complex environments, such as noise, illumination variations, and occlusion, during rocket booster recovery missions. Traditional Harris algorithms exhibit limitations in false detection rates, parameter sensitivity, and computational inefficiency. The proposed ARH integrates four key innovations: (1) Scharr operator-based gradient computation for enhanced edge response accuracy; (2) an adaptive thresholding mechanism derived from statistical analysis of Harris responses (), improving robustness against noise and dynamic lighting; (3) a hybrid approach combining dilation-accelerated non-maximum suppression (NMS) and sub-pixel refinement via OpenCV’sSgoodFeaturesToTrack, reducing redundancy and achieving sub-pixel localization (average displacement: 0.0458 pixels); and (4) vectorized computation replacing explicit loops, optimizing runtime efficiency by 40%. Experimental validation demonstrates ARH’s superior performance: high repeatability (≈0.85) under Gaussian noise, a 20% feature reduction rate in motion-blurred scenarios (vs. 21% for Harris and 82% for SIFT), and stable keypoint counts under rotation. Computational complexity analysis reveals that adaptive thresholding is optimized fromSStoS, while NMS efficiency improves by 50%. These advancements position ARH as a high-precision, real-time solution for vision-critical tasks in aerospace applications, such as rocket booster tracking and landing assistance.