Accurate Disparity Estimation Research Articles

OPAL: Occlusion Pattern Aware Loss for Unsupervised Light Field Disparity Estimation.

Light field disparity estimation is an essential task in computer vision. Currently, supervised learning-based methods have achieved better performance than both unsupervised and optimization-based methods. However, the generalization capacity of supervised methods on real-world data, where no ground truth is available for training, remains limited. In this paper, we argue that unsupervised methods can achieve not only much stronger generalization capacity on real-world data but also more accurate disparity estimation results on synthetic datasets. To fulfill this goal, we present the Occlusion Pattern Aware Loss, named OPAL, which successfully extracts and encodes general occlusion patterns inherent in the light field for calculating the disparity loss. OPAL enables: i) accurate and robust disparity estimation by teaching the network how to handle occlusions effectively and ii) significantly reduced network parameters required for accurate and efficient estimation. We further propose an EPI transformer and a gradient-based refinement module for achieving more accurate and pixel-aligned disparity estimation results. Extensive experiments demonstrate our method not only significantly improves the accuracy compared with SOTA unsupervised methods, but also possesses stronger generalization capacity on real-world data compared with SOTA supervised methods. Last but not least, the network training and inference efficiency are much higher than existing learning-based methods. Our code will be made publicly available.

Digital surface model generation from high‐resolution satellite stereos based on hybrid feature fusion network

AbstractRecent studies have demonstrated that deep learning‐based stereo matching methods (DLSMs) can far exceed conventional ones on most benchmark datasets by both improving visual performance and decreasing the mismatching rate. However, applying DLSMs on high‐resolution satellite stereos with broad image coverage and wide terrain variety is still challenging. First, the broad coverage of satellite stereos brings a wide disparity range, while DLSMs are limited to a narrow disparity range in most cases, resulting in incorrect disparity estimation in areas with contradictory disparity ranges. Second, high‐resolution satellite stereos always comprise various terrain types, which is more complicated than carefully prepared datasets. Thus, the performance of DLSMs on satellite stereos is unstable, especially for intractable regions such as texture‐less and occluded regions. Third, generating DSMs requires occlusion‐aware disparity maps, while traditional occlusion detection methods are not always applicable for DLSMs with continuous disparity. To tackle these problems, this paper proposes a novel DLSM‐based DSM generation workflow. The workflow comprises three steps: pre‐processing, disparity estimation and post‐processing. The pre‐processing step introduces low‐resolution terrain to shift unmatched disparity ranges into a fixed scope and crops satellite stereos to regular patches. The disparity estimation step proposes a hybrid feature fusion network (HF2Net) to improve the matching performance. In detail, HF2Net designs a cross‐scale feature extractor (CSF) and a multi‐scale cost filter. The feature extractor differentiates structural‐context features in complex scenes and thus enhances HF2Net's robustness to satellite stereos, especially on intractable regions. The cost filter filters out most matching errors to ensure accurate disparity estimation. The post‐processing step generates initial DSM patches with estimated disparity maps and then refines them for the final large‐scale DSMs. Primary experiments on the public US3D dataset showed better accuracy than state‐of‐the‐art methods, indicating HF2Net's superiority. We then created a self‐made Gaofen‐7 dataset to train HF2Net and conducted DSM generation experiments on two Gaofen‐7 stereos to further demonstrate the effectiveness and practical capability of the proposed workflow.

Bidirectional Semi-supervised Dual-branch CNN for Robust 3D Reconstruction of Stereo Endoscopic Images via Adaptive Cross and Parallel Supervisions.

Semi-supervised learning via teacher-student network can train a model effectively on a few labeled samples. It enables a student model to distill knowledge from the teacher's predictions of extra unlabeled data. However, such knowledge flow is typically unidirectional, having the accuracy vulnerable to the quality of teacher model. In this paper, we seek to robust 3D reconstruction of stereo endoscopic images by proposing a novel fashion of bidirectional learning between two learners, each of which can play both roles of teacher and student concurrently. Specifically, we introduce two self-supervisions, i.e., Adaptive Cross Supervision (ACS) and Adaptive Parallel Supervision (APS), to learn a dual-branch convolutional neural network. The two branches predict two different disparity probability distributions for the same position, and output their expectations as disparity values. The learned knowledge flows across branches along two directions: a cross direction (disparity guides distribution in ACS) and a parallel direction (disparity guides disparity in APS). Moreover, each branch also learns confidences to dynamically refine its provided supervisions. In ACS, the predicted disparity is softened into a unimodal distribution, and the lower the confidence, the smoother the distribution. In APS, the incorrect predictions are suppressed by lowering the weights of those with low confidence. With the adaptive bidirectional learning, the two branches enjoy well-tuned mutual supervisions, and eventually converge on a consistent and more accurate disparity estimation. The experimental results on four public datasets demonstrate our superior accuracy over other state-of-the-arts with a relative decrease of averaged disparity error by at least 9.76%.

Light-field spectral decomposition with a spatial–angular consistency prior for disparity estimation

Modeling the spatial–angular relationship is the theoretical basis for accurate disparity estimation from the light-field data. In this paper, the spatial–angular coupling relationship of light-field is established under the assumption of constant radiance along rays, and the spatial–angular consistency model of light-field is established under the assumption of Lambertian radiance. Based on the spatial–angular consistency prior of light-field, the light-field spectral decomposition based on disparity layers can be modeled as the ill-posed inverse problem. Based on the spectral decomposition, the disparity estimation is transformed into the process of searching for the spatial domain support set in the scene disparity hierarchy. The normalized cross-correlation function is utilized to estimate the disparity based on spectral decomposition under the spatial–angular consistency of the light-field. The performance on both simulated and real captured data show that the proposed algorithm is effective at high-precision and dense disparity estimation.

CGFNet: 3D Convolution Guided and Multi-scale Volume Fusion Network for fast and robust stereo matching

Nowadays, although significant progress has been made by convolutional neural network, it is still difficult to realize accurate and robust stereo matching in real time. In this article, we study how to achieve more accurate and robust disparity estimation based on real-time requirement. For this reason, a Multi-scale Volume Fusion (MVF) module was proposed and embedded to improve the matching accuracy. To achieve real-time performance, an innovative way to use 3D convolution is proposed. The 3D convolution is used during training for guidance and supervision, making the inference lightweight. Based on these two structures, we designed an end-to-end stereo matching method called 3D Convolution Guided and Multi-scale Cost Volume Fusion Network (CGFNet). Experimental results showed that our CGFNet has better generalization performance on cross-domain datasets, which achieves more accurate disparity estimation without additional fine tuning process in challenging regions. On KITTI benchmark, CGFNet reached D1-all=1.98% with substantial improvement among the State-Of-The-Art (SOTA) real-time models and runs a pair of images within 38 ms (26 fps). The results are notable when considering both matching accuracy and real-time performance.

Multi-scale graph neural network for global stereo matching

Currently, deep learning-based stereo matching is solely based on local convolution networks, which lack enough global information for accurate disparity estimation. Motivated by the excellent global representation of the graph, a novel Multi-scale Graph Neural Network (MGNN) is proposed to essentially improve stereo matching from the global aspect. Firstly, we construct the multi-scale graph structure, where the multi-scale nodes with projected multi-scale image features can be directly linked by the inner-scale and cross-scale edges, instead of solely relying on local convolutions for deep learning-based stereo matching. To enhance the spatial position information at non-Euclidean multi-scale graph space, we further propose a multi-scale position embedding to embed the potential position features of Euclidean space into projected multi-scale image features. Secondly, we propose the multi-scale graph feature inference to extract global context information on multi-scale graph structure. Thus, the features not only be globally inferred on each scale, but also can be interactively inferred across different scales to comprehensively consider global context information with multi-scale receptive fields. Finally, MGNN is deployed into dense stereo matching and experiments demonstrate that our method achieves state-of-the-art performance on Scene Flow, KITTI 2012/2015, and Middlebury Stereo Evaluation v.3/2021.

Adaptive Recurrent Iterative Updating Stereo Matching Network

When training a stereo matching network with a single training dataset, the network may overly rely on the learned features of the single training dataset due to differences in the training dataset scenes, resulting in poor performance on all datasets. Therefore, feature consistency between matched pixels is a key factor in solving the network’s generalization ability. To address this issue, this paper proposed a more widely applicable stereo matching network that introduced whitening loss into the feature extraction module of stereo matching, and significantly improved the applicability of the network model by constraining the variation between salient feature pixels. In addition, this paper used a GRU iterative update module in the disparity update calculation stage, which expanded the model’s receptive field at multiple resolutions, allowing for precise disparity estimation not only in rich texture areas but also in low texture areas. The model was trained only on the Scene Flow large-scale dataset, and the disparity estimation was conducted on mainstream datasets such as Middlebury, KITTI 2015, and ETH3D. Compared with earlier stereo matching algorithms, this method not only achieves more accurate disparity estimation but also has wider applicability and stronger robustness.

A Confidence-Aware Cascade Network for Multi-Scale Stereo Matching of Very-High-Resolution Remote Sensing Images

Stereomatching plays an essential role in 3D reconstruction using very-high-resolution (VHR) remote sensing images. However, it still faces unignorable challenges due to the multi-scale objects in large scenes and the multi-modality probability distribution in challenging regions, especially the occluded and textureless areas. Accurate disparity estimation in stereo matching for multi-scale objects has become a hard but crucial task. In this paper, to tackle these problems, we design a novel confidence-aware unimodal cascade and fusion pyramid network for stereo matching. The fused cost volume from the coarsest scale is used to generate the initial disparity map, and then the learnable confidence maps are generated to construct the unimodal cost distributions, which are used to narrow down the next-stage disparity search range. Moreover, we design a cross-scale interaction aggregation module to leverage multi-scale information. Both smooth-L1 loss and stereo focal loss are applied to regularize the disparity map and unimodal cost distribution, respectively. Compared to two state-of-the-art stereo matching networks, extensive experimental results show that our proposed network outperforms them in terms of average endpoint error (EPE) and the fraction of erroneous pixels (D1).

Sparse LIDAR Measurement Fusion with Joint Updating Cost for Fast Stereo Matching

The complementary virtues of active and passive depth sensors inspire the LIDAR-Stereo fusion for enhancing the accuracy of stereo matching. However, most of the fusion based stereo matching algorithms have exploited dense LIDAR priors with single fusion methodology. In this paper, we intend to break these fetters, utilizing sparse LIDAR priors with multi-step fusion strategy for obtaining accurate disparity estimation more efficiently. At first, random sparse sampling LIDAR depth measurements are provided in Naive Fusion for updating the matching cost ofSemi-Global Matching (SGM).Then Neighborhood Based Fusion is performed based on the former step for further updating the cost. Subsequently, Diffusion Based Fusion is utilized to update both the cost and disparities. At last, Tree Filtering is applied for removing speckle outliers and smoothing disparities. Performance evaluations on various stereo data sets demonstrate that the proposed algorithm outperforms other most challenging stereo matching algorithms significantly with approximately real-time implementation efficiency. Furthermore, it is worth pointing out that our proposal surprisingly possessesoneof the toptenperformances on Middleburyv.3online evaluation system even if it has not been adopted any learning-based techniques.

Deep Event Stereo Leveraged by Event-to-Image Translation

Depth estimation in real-world applications requires precise responses to fast motion and challenging lighting conditions. Event cameras use bio-inspired event-driven sensors that provide instantaneous and asynchronous information of pixel-level log intensity changes, which makes them suitable for depth estimation in such challenging conditions. However, as the event cameras primarily provide asynchronous and spatially sparse event data, it is hard to provide accurate dense disparity map in stereo event camera setups - especially in estimating disparities on local structures or edges. In this study, we develop a novel deep event stereo network that reconstructs spatial intensity image features from embedded event streams and leverages the event features using the reconstructed image features to compute dense disparity maps. To this end, we propose a novel event-to-image translation network with a cross-semantic attention mechanism that calculates the global semantic context of the event features for the intensity image reconstruction. In addition, a feature aggregation module is developed for accurate disparity estimation, which modulates the event features with the reconstructed image features by a stacked dilated spatially-adaptive denormalization mechanism. Experimental results reveal that our method can outperform the state-of-the-art methods by significant margins both in quantitative and qualitative measures.

Shape Prior Guided Instance Disparity Estimation for 3D Object Detection.

In this paper, we propose a novel system named Disp R-CNN for 3D object detection from stereo images. Many recent works solve this problem by first recovering point clouds with disparity estimation and then apply a 3D detector. The disparity map is computed for the entire image, which is costly and fails to leverage category-specific prior. In contrast, we design an instance disparity estimation network (iDispNet) that predicts disparity only for pixels on objects of interest and learns a category-specific shape prior for more accurate disparity estimation. To address the challenge from scarcity of disparity annotation in training, we propose to use a statistical shape model to generate dense disparity pseudo-ground-truth without the need of LiDAR point clouds, which makes our system more widely applicable. Experiments on the KITTI dataset show that, when LiDAR ground-truth is not used at training time, Disp R-CNN outperforms previous state-of-the-art methods based on stereo input by 20 percent in terms of average precision for all categories. The code and pseudo-ground-truth data are available at the project page: https://github.com/zju3dv/disprcnn.

ZoomNet: Part-Aware Adaptive Zooming Neural Network for 3D Object Detection

3D object detection is an essential task in autonomous driving and robotics. Though great progress has been made, challenges remain in estimating 3D pose for distant and occluded objects. In this paper, we present a novel framework named ZoomNet for stereo imagery-based 3D detection. The pipeline of ZoomNet begins with an ordinary 2D object detection model which is used to obtain pairs of left-right bounding boxes. To further exploit the abundant texture cues in rgb images for more accurate disparity estimation, we introduce a conceptually straight-forward module – adaptive zooming, which simultaneously resizes 2D instance bounding boxes to a unified resolution and adjusts the camera intrinsic parameters accordingly. In this way, we are able to estimate higher-quality disparity maps from the resized box images then construct dense point clouds for both nearby and distant objects. Moreover, we introduce to learn part locations as complementary features to improve the resistance against occlusion and put forward the 3D fitting score to better estimate the 3D detection quality. Extensive experiments on the popular KITTI 3D detection dataset indicate ZoomNet surpasses all previous state-of-the-art methods by large margins (improved by 9.4% on APbv (IoU=0.7) over pseudo-LiDAR). Ablation study also demonstrates that our adaptive zooming strategy brings an improvement of over 10% on AP3d (IoU=0.7). In addition, since the official KITTI benchmark lacks fine-grained annotations like pixel-wise part locations, we also present our KFG dataset by augmenting KITTI with detailed instance-wise annotations including pixel-wise part location, pixel-wise disparity, etc.. Both the KFG dataset and our codes will be publicly available at https://github.com/detectRecog/ZoomNet.

A Novel Stereo Matching Algorithm for Digital Surface Model (DSM) Generation in Water Areas

Image dense matching has become one of the widely used means for DSM generation due to its good performance in both accuracy and efficiency. However, for water areas, the most common ground object, accurate disparity estimation is always a challenge to excellent image dense matching methods, as represented by semi-global matching (SGM), due to the poor texture. For this reason, a great deal of manual editing is always inevitable before practical applications. The main reason for this is the lack of uniqueness of matching primitives, with fixed size and shape, used by those methods. In this paper, we propose a novel DSM generation method, namely semi-global and block matching (SGBM), to achieve accurate disparity and height estimation in water areas by adaptive block matching instead of pixel matching. First, the water blocks are extracted by seed point growth, and an adaptive block matching strategy considering geometrical deformations, called end-block matching (EBM), is adopted to achieve accurate disparity estimation. Then, the disparity of all other pixels beyond these water blocks is obtained by SGM. Last, the median value of height of all pixels within the same block is selected as the final height for this block after forward intersection. Experiments are conducted on ZiYuan-3 (ZY-3) stereo images, and the results show that DSM generated by our method in water areas has high accuracy and visual quality.

Fast and Accurate 3D Measurement Based on Light-Field Camera and Deep Learning.

The precise combination of image sensor and micro-lens array enables light-field cameras to record both angular and spatial information of incoming light, therefore, one can calculate disparity and depth from one single light-field image captured by one single light-field camera. In turn, 3D models of the recorded objects can be recovered, which means a 3D measurement system can be built using a light-field camera. However, reflective and texture-less areas in light-field images have complicated conditions, making it hard to correctly calculate disparity with existing algorithms. To tackle this problem, we introduce a novel end-to-end network VommaNet to retrieve multi-scale features from reflective and texture-less regions for accurate disparity estimation. Meanwhile, our network has achieved similar or better performance in other regions for both synthetic light-field images and real-world data compared to the state-of-the-art algorithms.

A Low-Cost Real-Time Embedded Stereo Vision System for Accurate Disparity Estimation Based on Guided Image Filtering

Stereo matching, a key element towards extracting depth information from stereo images, is widely used in several embedded consumer electronic and multimedia systems. Such systems demand high processing performance and accurate depth perception, while their deployment in embedded and mobile environments implies that cost, energy and memory overheads need to be minimized. Hardware acceleration has been demonstrated in efficient embedded stereo vision systems. To this end, this paper presents the design and implementation of a hardware-based stereo matching system able to provide high accuracy and concurrently high performance for embedded vision devices, which are associated with limited hardware and power budget. We first implemented a compact and efficient design of the guided image filter, an edge-preserving filter, which reduces the hardware complexity of the implemented stereo algorithm, while at the same time maintains high-quality results. The guided filter design is used in two parts of the stereo matching pipeline, showing that it can simplify the hardware complexity of the Adaptive Support Weight aggregation step, and efficiently enable a powerful disparity refinement unit, which improves matching accuracy, even though cost aggregation is based on simple, fixed support strategies. We implemented several variants of our design on a Kintex-7 FPGA board, which was able to process HD video (1,280 × 720) in real-time (60 fps), using ∼57.5k and ∼71k of the FPGA's logic (CLB) and register resources, respectively. Additionally, the proposed stereo matching design delivers leading accuracy when compared to state-of-the-art hardware implementations based on the Middlebury evaluation metrics (at least 1.5 percent less bad matching pixels).

Accurate disparity estimation in light field using ground control points

The recent development of light field cameras has received growing interest, as their rich angular information has potential benefits for many computer vision tasks. In this paper, we introduce a novel method to obtain a dense disparity map by use of ground control points (GCPs) in the light field. Previous work optimizes the disparity map by local estimation which includes both reliable points and unreliable points. To reduce the negative effect of the unreliable points, we predict the disparity at non-GCPs from GCPs. Our method performs more robustly in shadow areas than previous methods based on GCP work, since we combine color information and local disparity. Experiments and comparisons on a public dataset demonstrate the effectiveness of our proposed method.

Content-Based Guided Image Filtering, Weighted Semi-Global Optimization, and Efficient Disparity Refinement for Fast and Accurate Disparity Estimation

This paper presents a novel approach, which relies on content-based guided image filtering and weighted semi-global optimization for fast and accurate disparity estimation. The approach uses a pixel-based cost term that combines gradient, Gabor-Feature, and color information. The pixel-based matching costs are filtered by applying guided image filtering, which relies on rectangular support windows of two different sizes. In this way, two filtered costs are estimated for each pixel. Among the two filtered costs, the one that will be finally assigned to each pixel depends on the local image content around this pixel. The filtered cost volume is further refined by exploiting weighted semi-global optimization, which improves the disparity estimation accuracy. Finally, the disparity refinement in outlier regions relies on a straightforward and time-efficient outliers handling scheme and on a simple approach which deals with the disparity outliers at depth discontinuities. Experimental results on the Middlebury online stereo evaluation benchmark and 27 additional Middlebury stereo pairs prove that our method is able to generate disparity maps with high accuracy while keeping the computational cost low.

PM-PM: PatchMatch with Potts Model for object segmentation and stereo matching.

This paper presents a unified variational formulation for joint object segmentation and stereo matching, which takes both accuracy and efficiency into account. In our approach, depth-map consists of compact objects, each object is represented through three different aspects: 1) the perimeter in image space; 2) the slanted object depth plane; and 3) the planar bias, which is to add an additional level of detail on top of each object plane in order to model depth variations within an object. Compared with traditional high quality solving methods in low level, we use a convex formulation of the multilabel Potts Model with PatchMatch stereo techniques to generate depth-map at each image in object level and show that accurate multiple view reconstruction can be achieved with our formulation by means of induced homography without discretization or staircasing artifacts. Our model is formulated as an energy minimization that is optimized via a fast primal-dual algorithm, which can handle several hundred object depth segments efficiently. Performance evaluations in the Middlebury benchmark data sets show that our method outperforms the traditional integer-valued disparity strategy as well as the original PatchMatch algorithm and its variants in subpixel accurate disparity estimation. The proposed algorithm is also evaluated and shown to produce consistently good results for various real-world data sets (KITTI benchmark data sets and multiview benchmark data sets).

Cross-Based Local Stereo Matching Using Orthogonal Integral Images

We propose an area-based local stereo matching algorithm for accurate disparity estimation across all image regions. A well-known challenge to local stereo methods is to decide an appropriate support window for the pixel under consideration, adapting the window shape or the pixelwise support weight to the underlying scene structures. Our stereo method tackles this problem with two key contributions. First, for each anchor pixel an upright cross local support skeleton is adaptively constructed, with four varying arm lengths decided on color similarity and connectivity constraints. Second, given the local cross-decision results, we dynamically construct a shape-adaptive full support region on the fly, merging horizontal segments of the crosses in the vertical neighborhood. Approximating image structures accurately, the proposed method is among the best performing local stereo methods according to the benchmark Middlebury stereo evaluation. Additionally, it reduces memory consumption significantly thanks to our compact local cross representation. To accelerate matching cost aggregation performed in an arbitrarily shaped 2-D region, we also propose an orthogonal integral image technique, yielding a speedup factor of 5-15 over the straightforward integration.