There is disclosed a depth information generation apparatus capable of acquiring high-precision depth information within a short computation time. A depth information generation apparatus of this invention generates depth information at the capture position of a reference image from the reference image, and at least one peripheral image that forms a stereo image pair with the reference image, and has a high-speed stereo processor (30) for generating depth information at high speed from the reference image and peripheral image, and a high-precision stereo processor (40) for generating high-precision depth information. A motion detector (20) detects a motion in an image, and instructs an image composition unit (50) to select the output from the high-speed stereo processor (30) for a portion with a large motion, and the output from the high-precision stereo processor (40) for other portions. The image composition unit (50) composites the outputs from the high-speed stereo processor (30) and high-precision stereo processor (40) in accordance with an instruction from the motion detector (20), and outputs a depth map as final depth information.