结合物体先验和空域约束的室内空域布局推理
发布时间：2018-09-10 11:40
【摘要】：对结构化室内场景的空域布局结构进行估计是计算机视觉领域的研究热点之一.然而,对于内部堆放了众多杂乱物体的室内场景,现有的大多数方法容易受到各种物体遮挡的影响而无法对这一类场景的布局结构进行准确推理.为此,本文方法充分考虑了房间和物体之间的几何和语义关联性,参数化地对房间和内部物体的三维体积分别进行描述,并且提出利用多种高层图像语义来获取物体的先验信息.此外,还在此基础上加入了空域排他性和空域位置等多种空域约束,进而在改进室内场景空域布局估计的同时为物体的识别和定位提供关键信息.本文方法不仅具有较低的求解复杂度,而且通过试验表明相比于现有的经典方法在杂乱的室内场景中能够取得更为鲁棒的空域布局推理结果.
[Abstract]:Estimating the spatial layout of structured interior scene is one of the hot topics in the field of computer vision. However most of the existing methods are vulnerable to the influence of various objects and can not infer the layout structure of this kind of scene accurately. In this paper, the geometric and semantic correlations between the room and the object are fully considered, and the three-dimensional volume of the room and the internal object are parameterized respectively. And a variety of high-level image semantics are proposed to obtain the prior information of objects. In addition, a variety of airspace constraints, such as spatial exclusivity and spatial location, are added to improve the spatial layout estimation of indoor scenes and provide key information for object recognition and location. The proposed method not only has a lower complexity, but also shows that compared with the existing classical methods, it can obtain more robust spatial layout reasoning results than the existing classical methods in cluttered indoor scenes.
【作者单位】： 宁波工程学院电信学院;浙江大学信息与电子工程学院;
【基金】：浙江省自然科学基金(LQ15F020004) 浙江省公益类技术研究项目(2016C33255) 宁波市自然科学基金(2015A610132,2013A610113)资助~~
【分类号】：TP391.41
，
本文编号：2234329
[Abstract]:Estimating the spatial layout of structured interior scene is one of the hot topics in the field of computer vision. However most of the existing methods are vulnerable to the influence of various objects and can not infer the layout structure of this kind of scene accurately. In this paper, the geometric and semantic correlations between the room and the object are fully considered, and the three-dimensional volume of the room and the internal object are parameterized respectively. And a variety of high-level image semantics are proposed to obtain the prior information of objects. In addition, a variety of airspace constraints, such as spatial exclusivity and spatial location, are added to improve the spatial layout estimation of indoor scenes and provide key information for object recognition and location. The proposed method not only has a lower complexity, but also shows that compared with the existing classical methods, it can obtain more robust spatial layout reasoning results than the existing classical methods in cluttered indoor scenes.
【作者单位】： 宁波工程学院电信学院;浙江大学信息与电子工程学院;
【基金】：浙江省自然科学基金(LQ15F020004) 浙江省公益类技术研究项目(2016C33255) 宁波市自然科学基金(2015A610132,2013A610113)资助~~
【分类号】：TP391.41
，
本文编号：2234329
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2234329.html
