Aiming at the problem that the reconnaissance robot has poor immediacy in the real-time positioning and global map construction process because of the slow selection of key frames in the RGB-D SLAM algorithm,this paper proposes a SLAM key frame adaptive selection algorithm suitable for RGB-D cameras.The inter-frame similarity of the continuous images acquired by the camera may be changed due to the different motion states of the robot, which will affect the selection of key frames.In this paper, the neural network is used to construct the relationship model between the motion state of the robot and the step length of the key frame.The key frame adaptive selection algorithm based on the motion state of the robot is proposed to realize the adaptive selection of key frames.The experimental results show that, the algorithm reduces the amount of calculation of the system and improves the overall operating efficiency of the system.Therefore,the algorithm can effectively improve the running speed of the system while ensuring the accuracy and has strong practicability.