RGBD saliency object detection method based on 3D convolutional neural network

• Main Authors: CHEN QIAN, FU KEREN, ZHAO QIJUN
• Format: Patent
• Language: Chinese; English
• Published: 04.05.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention discloses an RGBD saliency object detection method based on a 3D convolutional neural network, and the method comprises the steps: obtaining an RGB image and a depth image of a to-be-detected image, converting the depth image into three channels, carrying out the series connection of the RGB image and the depth image of the three channels in a time dimension, outputting a 4D tensor with the time dimension, inputting the 4D tensor into a 3D encoder in the 3D convolutional neural network, outputting a series of side path hierarchical features, wherein the 3D encoder is a residual network after time dimension expansion; and a 3D decoder in the 3D convolutional neural network receives the hierarchical features, performs compression, recursive decoding and excitation on the hierarchical features, and finally outputs RGBD saliency object detection results, and the series of side path hierarchical features are connected in series in the time dimension during recursive decoding. According to the inventi