<p>This paper proposes a feature point extraction method based on grid partitioning and adaptive threshold adjustment, aiming to enhance the stability of feature point extraction and the accuracy of feature matching under varying illumination conditions. The method employs a B-Spline image pyramid to improve the detail preservation capability of multi-scale images, and integrates a QuadTree-based feature point uniformization technique to optimize the distribution of feature points, thereby avoiding redundant extraction in dense regions. Moreover, this paper introduces a grid-based motion statistical feature matching algorithm that adaptively adjusts the matching threshold by analyzing the motion characteristics within each grid, thus improving matching accuracy and robustness in complex environments. The proposed method is evaluated on the Oxford Affine dataset, and experimental results demonstrate that, compared with the traditional ORB algorithm, the method can extract more robust feature points under various conditions such as blur, illumination changes, and compression, thereby enhancing the accuracy of matching. In summary, this method provides an effective solution for feature extraction and matching in complex environments.</p>