Object Recognition and Pose Estimation base on Deep Learning

In this paper, a deep learning method is proposed to solve the problem of second pose determination of hand-eye system. The pose information of the first image is introduced by the convolution neural network, and the main module of the pose estimation of the first image is formed into a joint deep learning framework. Through human supervision, building a set of training samples, through these training samples to optimize the parameters of the model, and through the first piece of practical image, complete the estimation of the first image uptake posture, and then draw the robot second images uptake pose. By analyzing the disparity between the second image and the first one, it is found that the maximum range of the disparity map contains all the feature parallax. The model can effectively estimate the second pose of the robot by using the object pose information of the first image, and obtain a set of the best 3D matching images.