Traitement en cours

Veuillez attendre...

Paramétrages

Paramétrages

Aller à Demande

1. CN112435282 - Real-time binocular stereo matching method based on adaptive candidate parallax prediction network

Office
Chine
Numéro de la demande 202011176728.1
Date de la demande 28.10.2020
Numéro de publication 112435282
Date de publication 02.03.2021
Type de publication A
CIB
G06T 7/33
GPHYSIQUE
06CALCUL; COMPTAGE
TTRAITEMENT OU GÉNÉRATION DE DONNÉES D'IMAGE, EN GÉNÉRAL
7Analyse d'image
30Détermination des paramètres de transformation pour l'alignement des images, c. à d. recalage des images
33utilisant des procédés basés sur les caractéristiques
G06T 7/593
GPHYSIQUE
06CALCUL; COMPTAGE
TTRAITEMENT OU GÉNÉRATION DE DONNÉES D'IMAGE, EN GÉNÉRAL
7Analyse d'image
50Récupération de la profondeur ou de la forme
55à partir de plusieurs images
593à partir d’images stéréo
G06N 3/04
GPHYSIQUE
06CALCUL; COMPTAGE
NSYSTÈMES DE CALCULATEURS BASÉS SUR DES MODÈLES DE CALCUL SPÉCIFIQUES
3Systèmes de calculateurs basés sur des modèles biologiques
02utilisant des modèles de réseaux neuronaux
04Architecture, p.ex. topologie d'interconnexion
CPC
G06T 7/33
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
7Image analysis
30Determination of transform parameters for the alignment of images, i.e. image registration
33using feature-based methods
G06T 7/593
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
7Image analysis
50Depth or shape recovery
55from multiple images
593from stereo images
G06N 3/0454
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
04Architectures, e.g. interconnection topology
0454using a combination of multiple neural nets
G06T 2207/20228
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
2207Indexing scheme for image analysis or image enhancement
20Special algorithmic details
20228Disparity calculation for image-based rendering
G06T 2207/10012
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
2207Indexing scheme for image analysis or image enhancement
10Image acquisition modality
10004Still image; Photographic image
10012Stereo images
G06T 2207/20081
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
2207Indexing scheme for image analysis or image enhancement
20Special algorithmic details
20081Training; Learning
Déposants XI'AN JIAOTONG UNIVERSITY
西安交通大学
Inventeurs ZHANG XUCHONG
张旭翀
SUN HONGBIN
孙宏滨
DAI HE
戴赫
ZHAO YONGLI
赵永利
ZHENG NANNING
郑南宁
Mandataires 西安通大专利代理有限责任公司 61200
Titre
(EN) Real-time binocular stereo matching method based on adaptive candidate parallax prediction network
(ZH) 一种基于自适应候选视差预测网络的实时双目立体匹配方法
Abrégé
(EN) The invention discloses a real-time binocular stereo matching method based on an adaptive candidate parallax prediction network. The method comprises the following steps: firstly, carrying out multi-scale feature extraction on a three-dimensional image pair by utilizing a two-dimensional convolutional neural network to obtain high-resolution and low-resolution feature maps; then, in the first stage of parallax estimation, using a low-resolution feature map is used to perform parallax coarse estimation; after the adaptive candidate parallax is predicted by using the rough estimation result andthe left image information, in the second-stage parallax estimation, carrying out fine parallax estimation by using the prediction result and the high-resolution feature map; and finally, performing hierarchical refinement on the disparity map to obtain a full-size disparity map. Compared with an existing coarse-to-fine stereo matching neural network, the invention has the advantages that more accurate dynamic offset can be predicted for a fine parallax estimation stage so as to meet different parallax correction requirements of various targets in the image. Due to the effectiveness of dynamicprediction, a two-stage processing structure is designed to greatly improve the calculation precision and speed of the real-time binocular stereo matching network.
(ZH) 本发明公开了一种基于自适应候选视差预测网络的实时双目立体匹配方法。该方法首先利用二维卷积神经网络对立体图像对进行多尺度特征提取,得到高、低分辨率的特征图。然后,第一阶段视差估计利用低分辨率特征图进行视差粗估计。在利用粗估计结果和左图信息预测得到自适应候选视差以后,第二阶段视差估计则利用预测结果和高分辨率特征图进行精细视差估计。最后,对视差图进行层次化精修得到全尺寸视差图。与现有的由粗到精立体匹配神经网络相比,本发明可以为精细视差估计阶段预测更准确的动态偏移量,以满足图像中各种目标不同的视差校正需求。由于动态预测的有效性,本发明设计了两级处理结构以大幅提高实时双目立体匹配网络的计算精度和速度。
Documents de brevet associés