The color and depth information of multi-scenario, multi-vision and multiple target in the RGB-D images are conveniently obtained using a new 3D sensor at the same time. The RGB-D image classification accuracy is effectively improved using the depth information invariant characteristics of color and brightness, when the objects overlap and occlusion occurs. The development and theory of Microsoft Kinect are introduced in detail, and the existing RGB-D datasets are described. Then the feature extraction and classification methods are summarized, analyzed and compared. The development trend of RGB-D image classification is discussed.