Pose estimation is one of the most popular research directions in the field of intelligent and automatic control at present, and it has many applications in the fields of unmanned vehicle, intelligent robotic arm, and household robot. However, most of the traditional methods are complicated in calculation and difficult in real-time computing. A pose calculation scheme is proposed, which uses convolution neural network to compress the size and extract the information of the image input of binocular camera, and carries out regression learning on the feature vectors through the bidirectional long-short time neural network and the standard results calculated by lidar. The trained deep learning solution has a certain improvement in accuracy and speed compared with the traditional solution.