The power switch is a basic component of the power system. The switch status during system operation and maintenance often needs to be manually confirmed for multiple times. Under the background, a new method for recognizing the state of the knife switch is proposed. Firstly, the input 3D knife gate image is converted into color point cloud data, and the scene features are extracted from the color point cloud data. Then, the existing direction histogram color feature description algorithm is employed to construct the recognition feature of descriptor based on the local texture and shape information. With the given extracted feature descriptors, a two-stage matching process is performed to find the correspondence between the scene and the color point cloud model of the target. Secondly, the Hough voting algorithm is utilized to filter the matching errors in the corresponding set and estimate the initial three-dimensional posture of the knife gate. Finally, in the pose estimation stage, the random sample consistency and hypothesis verification algorithms are used to improve the initial pose and filter out bad estimation results with incorrect assumptions. The experimental results show that the method can successfully identify the switch parts in complex power scenes, and can accurately estimate the three-dimensional posture information of the target.