Aiming at the interactive requirements of the virtual display device of phantom imaging in museums,a computer vision interactive system with natural gestures is designed.The monocular camera is used to collect the action of natural gestures.The optimal capture position of the monocular camera is calculated according to the range of hand activities of people with different heights,and the gesture recognition of 140～190 cm crowd height is realized.The Mediapipe machine learning framework is used to traverse the captured gesture image in real time to obtain the calibration position of a single hand frame.Combined with the palm model of 21 feature points,the non-maximum suppression algorithm is used to identify the self-occluded palm.According to the Euclidean space distance discrimination threshold and the curvature of a single finger,the inter-finger actions are classified.Five common interactive actions in phantom imaging are defined,and the real-time mapping between the fingertip and the model feature points is established through the coordinate relationship.The experimental results show that the recognition accuracy of the interactive system designed in this paper reaches 98%,which meets the requirements of gesture control virtual model in phantom imaging system.