Article
Computer Science, Artificial Intelligence
Aixin Chong, Hui Yin, Yanting Liu, Jin Wan, Zhihao Liu, Ming Han
Summary: Compared with traditional hand-crafted feature based methods, learning-based stereo matching methods have made significant progress in matching accuracy. However, current CNN-based methods often require a substantial amount of time and memory consumption. To address this issue, we propose an accurate and fast stereo matching network that incorporates multi-hierarchy feature extraction and multi-step cost aggregation. Experimental results demonstrate that our network achieves highly competitive disparity estimation accuracy with fast inference speed.
Article
Computer Science, Artificial Intelligence
Hamid Laga, Laurent Valentin Jospin, Farid Boussaid, Mohammed Bennamoun
Summary: This paper provides a comprehensive survey of deep learning-based stereo depth estimation, summarizing the commonly used methods and discussing their advantages and limitations. The growing success of deep learning in solving 2D and 3D vision problems has attracted increasing research interest in this field, leading to significant performance improvement.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Information Systems
Yong Deng, Jimin Xiao, Steven Zhiying Zhou
Summary: In this work, an end-to-end Time-of-Flight (ToF) and stereo data fusion network is proposed, which incorporates prior ToF depth knowledge into the stereo matching process to achieve more accurate depth maps. The dynamic search range for each pixel is determined based on an estimated ToF error map, which improves efficiency and effectiveness in handling different errors. Experimental results demonstrate that the fusion method outperforms both ToF and stereo alone, as well as other state-of-the-art fusion methods on synthetic and real data.
IEEE TRANSACTIONS ON MULTIMEDIA
(2022)
Article
Engineering, Civil
Kai Zeng, Yaonan Wang, Qing Zhu, Jianxu Mao, Hui Zhang
Summary: This paper proposes a progressive fusion stereo matching network for depth estimation of rectified image pairs. It utilizes an encoder-decoder feature extraction network architecture and a multi-scale cost aggregation strategy. Experimental results demonstrate its superiority over previous methods.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Zhengfa Liang, Yulan Guo, Yiliu Feng, Wei Chen, Linbo Qiao, Li Zhou, Jianfeng Zhang, Hengzhu Liu
Summary: In this paper, an end-to-end trainable convolutional neural network is proposed for stereo matching, with shared feature extraction, initial disparity estimation, and disparity refinement as its three sub-modules, achieving good performance across multiple datasets.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Information Systems
Junda Cheng, Xin Yang, Yuechuan Pu, Peng Guo
Summary: This paper proposes a novel region separable stereo matching method which improves the performance of convolutional neural networks in stereo matching. The method automatically groups image pixels into regions and constructs and processes cost volumes for each region, leading to more accurate and efficient stereo matching. Experimental results demonstrate that the method significantly improves existing models.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Information Systems
Jaecheol Jeong, Suyeon Jeon, Yong Seok Heo
Summary: Recent stereo matching networks, although effective in accuracy, have limitations due to high computing and memory requirements. To address this, a Sequential Feature Fusion Network (SFFNet) was proposed, using 2D convolutions for cost volume generation, showing advantages in accuracy and efficiency over existing networks.
Article
Multidisciplinary Sciences
Chen-Wei Huang, Jian-Jiun Ding
Summary: The paper aims to address the limitations of conventional keypoint-based disparity estimation methods. A superpixel-based algorithm is proposed that utilizes both keypoint and semiglobal information to improve the accuracy of disparity estimation, especially in smooth and symmetric regions. A disparity refining mechanism is also applied to correct the disparity of superpixels with no or few keypoints, solving the problem of non-uniform distribution.
Article
Computer Science, Artificial Intelligence
Yeongmin Lee, Chong-Min Kyung
Summary: The research presents a memory-efficient and robust stereo matching algorithm using the semiglobal parametric approach and Gaussian mixture model to reduce memory bandwidth. Furthermore, a learning-based confidence measure is introduced through training features with the random forest framework, enhancing the overall performance. Experimental results demonstrate that the method significantly reduces memory requirements while maintaining robust depth map compared to traditional approaches.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Review
Environmental Sciences
Guobiao Yao, Alper Yilmaz, Fei Meng, Li Zhang
Summary: Researchers have been focusing on automatic matching for optical wide-baseline stereo images, with deep convolutional neural networks and learning-based methods showing potential over handcrafted features. By conducting comprehensive experiments and summarizing current representative research, they have provided insights and guidance for future work.
Article
Geochemistry & Geophysics
Vahid Mousavi, Masood Varshosaz, Fabio Remondino, Saied Pirasteh, Jonathan Li
Summary: In this article, a universal two-step filtering method is proposed to solve the keypoint matching problem in the presence of repetitive patterns. The method uses a mean-shift clustering algorithm and a novel confusion index to filter out mismatches. Experimental results show that the proposed strategy improves the accuracy of matching and photogrammetric blocks.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Hao Wang, Xiaolei Lv, Shuo Li
Summary: This study proposes a novel algorithm for building change detection that can process images from different perspectives and utilizes cross-temporal stereo matching to correct elevation errors in the existing DSM. This improves the accuracy and completeness of building change detection.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Engineering, Multidisciplinary
Yukun Han, Chong Pan, Zepeng Cheng, Yang Xu
Summary: The paper proposes a feature-point matching algorithm based on particle tracking velocimetry for measuring complex surface morphology. By mixing the epipolar-line constraint with a global similarity pairing, the algorithm iteratively estimates the depth of each feature point. Experimental results demonstrate the algorithm's superior accuracy and robustness compared to traditional methods.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2023)
Article
Computer Science, Software Engineering
Weijie Bi, Ming Chen, Dongliu Wu, Shenglian Lu
Summary: This paper proposes a new weighted smooth L1 loss function that considers the loss function calculation on edge regions to improve accuracy. It also introduces an improved bilateral grid upsampling module and a strategy to balance computational consumption for real-time inference. Experimental results demonstrate the simplicity and effectiveness of this approach, achieving an endpoint error improvement to 0.63 under 33 frames per second (FPS). Additionally, the proposed edge-based loss function can be easily embedded into existing stereo matching networks, such as GwcNet, AANet, and PSMNet, significantly reducing their endpoint errors.
Article
Chemistry, Analytical
Junhui Mei, Xiao Yang, Zhenxin Wang, Xiaobo Chen, Juntong Xi
Summary: This paper proposes a topology-based stereo matching method for 3D measurement using a single pattern of coded spot-array structured light. Experimental results demonstrate that the proposed technique can successfully decode each spot and establish stereo matching relation, achieving the goal of 3D reconstruction.