Abstract: A feature point matching method for complex environments is proposed to address the problem that traditional feature point extraction is affected by changes in illumination, viewing angle and image noise, resulting in poor accuracy of feature point tracking in subsequent optical flow algorithms. Inspired by the strong robustness of the SuperPoint network in feature extraction, a Hessian matrix was constructed on this basis to re-screen the detected feature points. For the existence of redundant descriptor information in the semi-dense descriptors of SuperPoint, the traditional BRIEF descriptors were proposed to be used instead to extract the screened feature points, and the BruteForce matching method was used for feature matching. The experimental results show that this method can effectively mitigate the effects of illumination, view angle and noise changes on feature point extraction, and can obtain better feature matching results.