Aiming at the problem of the large amount of data and the complicated algorithm in 3D scene feature recognition process, an feature extraction and reconstruction of environmental plane algorithm based on Kinect was proposed. Firstly, a method of RANdom SAmple Consensus (RANSAC) environment segmentation with the combination of geometrical and color information was proposed, which overcame the over segmentation and lack-segmentation based on geometric characteristic and improved the accuracy. Secondly, according to the principle of perspective projection, the three-dimensional transformation matrix was derived, which guided the 3D environment mapped into a plane. The extraction of contour points in two-dimensional space was realized by searching object boundary information which used convex hull concept. Finally, the 3D information of the contour points was recovered by the rotating inverse transform and the reconstruction of environment features was completed. Three groups of scene data were used to verify the algorithm and the experimental results show the proposed algorithm gains more precise segmentation, reduces over segmentation phenomena, and also has better reconstruction effect for objects with different shape features.