%0 Journal Article
%T 基于预测一致性嵌入的注视目标检测<br>Gaze Target Detection Based on Predictive Consistency Embedding
%A 史俊彪
%A 骆文杰
%A 熊思璇
%A 单东风
%A 江朝晖
%A 韩超
%J Journal of Image and Signal Processing
%P 144-157
%@ 2325-6745
%D 2023
%I Hans Publishing
%R 10.12677/JISP.2023.122015
%X 本文研究了第三人称视角下图像的注视目标检测问题我们提出了一个深度架构推断场景中的人在看哪里。该模型在蕴含丰富上下文信息的场景图像、深度图像和头部图像上进行训练。与现有的技术不同，我们的模型不需要监视注视角度，不依赖于头部方向信息和眼睛信息。大量的实验表明，我们的方法在多个基准数据集上具有更强的性能。我们还研究了注视目标检测的域自适应方法，使用一致性嵌入确保源域和目标域对齐，使得我们的模型能够有效地处理数据集之间的间隙。<br />
In this paper, we study the problem of gaze target detection in images from the third person perspective. We propose a deep architecture to infer where people are looking in the scene. The model is trained on scene image, depth image and head image containing rich contextual information. Unlike existing technologies, our model does not need to monitor gaze angles and does not rely on head direction information and eye information. A large number of experiments show that our method has stronger performance on multiple benchmark data sets. We also study a domain adaptive approach to gaze target detection, using consistency embedding to ensure the alignment of source and target domains, so that our model can effectively deal with gaps between datasets.
%K 注视目标检测，注视跟随，域自适应，RGB图像，深度图像<br>Gaze Target Detection
%K Gaze Follow
%K Domain Adaptation
%K RGB Image
%K Depth Image
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=64836