Vision-based Scene Recognition for Product Search
Vision-based Scene Recognition for Product Search
复制标题
用于产品搜索的基于视觉的场景识别
DOI：
10.1109/iceeie52663.2021.9616880
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Handayani Anik Nur
中科院分区：
文献类型：
--
作者：
Takayanagi Miho;Fukuda Osamu;Yamaguchi Nobuhiko;Okumura Hiroshi;Handayani Anik Nur
本文提出了一种将关联模型与一般目标识别相结合的方法。该方法能够检测出一般的目标，并能识别出场景。基于场景识别，可以提高目标检测的准确率，减少在商店搜索产品的时间。为了从检测到的目标中识别出场景，该算法使用了贝叶斯网络构建的相关性模型。通过实验验证了该方法的有效性。在实验中，估计了产品的库存状态和边角的类别。结果证实，对产品和边角品类的库存状况进行了正确估计。
This paper proposes a method that combines a relevance model with general object recognition. This method can detect general objects and recognize scenes. Based on scene recognition, object detection accuracy can be improved, and the product search time at a store can be reduced. To recognize a scene from the detected objects, the proposed algorithm uses a relevance model constructed by a Bayesian network. Experiments were conducted to verify the effectiveness of the proposed method. In the experiments, the stock status of the products and the categories of the corners were estimated. The results confirmed that the stock status of the products and corner categories were correctly estimated.