Method for detecting RGB-D (red, green and blue-depth) three-dimensional objects on basis of deep learning

The invention discloses a method for detecting RGB-D (red, green and blue-depth) three-dimensional objects on the basis of deep learning. The method includes labeling RGB-D images and acquiring labeled RGB-D image data sets to be used as training samples and test samples for three-dimensional object detection convolutional neural network models; building the three-dimensional object detection convolutional neural network models, and inputting the training samples and the test samples into the three-dimensional object detection convolutional neural network models; setting hyper-parameters of the three-dimensional object detection convolutional neural network models, training the three-dimensional object detection convolutional neural network models by the aid of Caffe and generating training models when cost loss functions are reduced to reach ideal degrees and the three-dimensional object detection convolutional neural network models are trained by required maximum numbers of iteration; inputting the RGB-D im