The ORB (Oriented FAST and Rotated BRIEF) feature detection algorithm often encounters challenges in envi-ronments with blur and drastic lighting variations, leading to significant disparities in the number of extracted features and the matching accuracy. Moreover, it tends to generate feature point clusters, particularly at image object corners. To address these issues, this paper proposes an improved ORB feature detection algorithm . Firstly, the Multi-Scale Retinex (MSR) algorithm is employed to enhance image features. Next, the image is divided into a grid, and thresholds for feature point detection are adjusted based on the grayscale distribution within each grid. Sub-sequently, a dynamic region-based non-maximum suppression method is applied to select the best feature points. Experimental results demonstrate that the algorithm, as improved in this paper, results in a more evenly distributed feature point layout on the image. In scenarios with lighting variations within an 80% range, the repeatability rate of feature points remains stable at over 75%, and the average matching accuracy improves by 22% compared to the original ORB algorithm.