The invention provides a feature matching method for binocular image splicing of a mobile inspection robot, and the method comprises the steps: constructing a scale pyramid in a feature point detection stage, extracting feature points through a FAST algorithm with an extremely high speed, and enhancing the robustness of scale invariance; adopting the improved CS -LBP description method to describe the feature points, enhancing robustness of rotation invariance, reducing the dimension of the feature vector, and improving the matching efficiency; and finally, measuring the similarity of the feature vectors by using a DDRN algorithm to complete matching, and eliminating mismatching by using an improved RANSAC algorithm. Compared with a traditional algorithm, the algorithm has the advantages that the real-time performance is greatly improved, meanwhile, feature extraction and matching of the image are accurately achieved, the improved description method is high in anti-interference performance on the rotating image, and high adaptability is still achieved in complex transformation scenes such as affine, zooming and illumination.