Fast recognizing, positioning and surface detection of multi-shaped objects in complex environment are studied to satisfy the requirement of smart machines, which is expected to grab the objects or inspect surface defection in complex environment in real time. Fast recognition, positioning, stereo matching and post-processing algorithm of point clouds are discussed. At first, new targets in the scene are recognized by robust principal component analysis, and the image location of the targets is accurately acquired by improved k-means clustering algorithm. Then, the region of interest is screened out by support vector machine, and one-dimensional searching is carried out by epipolar restriction to obtain the regions to be matched in binocular images, and local three-dimensional point clouds are quickly obtained. Finally, special denoising operation of point clouds is conducted to reduce the error. The experiment results indicate that the proposed algorithm effectively reduces the running time of the process and effectively reduces all the noises caused by complex backgrounds, and improves the accuracy and adaptability of point clouds acquisition in complex environment, and it is a robust, effective and fast algorithm for three-dimensional point clouds acquisition.