Depth map acquisition, which is based on laser speckle, presents some issues, such as low matching precision, large amount of calculation, and poor robustness in different measurement environments. In this paper, a semi-dense depth map acquisition algorithm based on laser speckle is proposed to address these issues. The problem of poor robustness can be solved using the locally adaptive binarization, which preprocesses the speckle map to ensure the illumination invariance of the window descriptor. In terms of measurement accuracy, the central pixel coordinates of each speckle are extracted using a clustering algorithm, which improves the positional accuracy of each speckle. Regarding the matching success rate issue, the window descriptor is convoluted to obtain a simplified descriptor, which is able to reduce the amount of calculations and increase the matching success rate. Finally, the speckle pairing points are obtained according to the matching criterion, and then the depth values of each speckle are obtained according to the triangulation principle. Experiments confirm that the proposed algorithm is highly robust and accurate and improves the matching success rate.