VR/AR全景图像/视频的显著性预测能够有效减少信息冗余，增强用户舒适感。但是当前对AR全景图像/视频的显著性预测研究仍是一项空白，对VR全景图像/视频的显著性预测方法研究大多聚焦于构建可计算的视觉注意模型，而忽略了类脑视觉注意机制。鉴于此，本项目拟开展VR/AR全景图像/视频中基于类脑视觉注意机制的显著性预测模型研究，具体包括三个方面：（1）探索类脑视觉注意感知融合机理，构建基于注意力感知特征融合的VR全景图像的显著性预测模型；（2）模拟类脑视觉注意机制，构建基于VR全景图像的浅层循环结构神经网络的轻量级显著性预测模型；（3）探索AR环境中特有的视觉注意机制，构建基于感受野模块网络的AR全景视频的轻量级显著性预测模型。以上每部分研究内容均遵循自底向上的“机理-模型-应用”一体化研究思路，其研究成果不仅可以推动新一代“脑启发人工智能”的发展，而且也促进VR/AR产品进入体验“舒适阶段”。

The saliency prediction of VR (Virtual Reality)/AR (Augmented Reality) panoramic images/videos can effectively reduce information redundancy and enhance user comfort. However, the current research on saliency prediction of AR panoramic images/videos is still blank. Most of the research on saliency prediction methods of VR panoramic images/videos focuses on the construction of computable visual attention models, and ignores brain-like visual attention mechanisms. Considering this, this project intends to conduct research on saliency prediction models based on brain-like visual attention mechanisms in VR/AR panoramic images/videos. Specifically, we plan to carry out research from the following three aspects. (1) We explore the mechanism of brain-like visual attention perception fusion, and construct saliency prediction model of VR panoramic image based on attention perception features fusion. (2) We simulate brain-like visual attention mechanism and propose a lightweight saliency prediction model of shallow recurrent structure neural network based on VR panoramic images. (3) We explore the unique the visual attention mechanism in the AR environment and construct a lightweight saliency prediction model of AR panoramic video based on the receptive field module network. The above three aspects of research contents follow a bottom-up and unified “mechanism-model-application” research route. The research results can not only promote the development of a new generation of “brain-inspired artificial intelligence”, but also push VR/AR products into the “comfort phase” of experience.