Computer Image Feature Extraction Algorithm for Robot Visual Equipment

To improve the poor condition of the robot image feature extraction with change in the environment, this thesis made improvement in the Local Binary Patterns (LBP), and proposed the SIFT-MLBP image feature extraction, coding by partition grid based on the correlation between the neighboring image center pixel points. After obtaining the key image characteristics by using the SIFT algorithm, a gridding structure was built centered on the pixel point in each region, calculating the partial difference between the pixel points, and assigning weight to each pixel encoding with different contrasts. In this study, the model-based feature vector combining the Gabor algorithm was extracted to build the SIFT-GMLBP feature vector, which reduced feature dimensions by mapping of the original complement with each other. The test showed that the SIFT-GMLBP algorithm possesses a fairly good feature matching effect, with the correct matching rate over 95%, and reduced running time of 0.05S. The robustness of this method in dealing with the external environment is quite remarkable, as it is able to improve the speed and precision of the mobile robots’ image identification in the complex environment.