Abstract: For the incremental computation and real-time problems of the feature extraction in the self-learning process of developmental robot, an incremental BDPCA (bidirectional principal component analysis) algorithm based on CCIPCA (candid covariance-free incremental principal component analysis) and BDPCA algorithms is proposed. The iterative calculation method is also adopted with the incremental computation ability. In the proposed algorithm, the 2-dimensional original image matrix is taken as the processing object directly, which effectively reduces the computation cost and shortens the running time. To verify the proposed algorithm, the support vector machine method is used to classify the building blocks grasped by the manipulator. The experimental results show that the algorithm is effective and can increase the average classification rate to 90%. The processing speed is approximately 26 frames per second, which can meet the real-time processing needs of developmental robots.