Depth Map Fusion Research Articles

LapUNet: a novel approach to monocular depth estimation using dynamic laplacian residual U-shape networks

Monocular depth estimation is an important but challenging task. Although the performance has been improved by adopting various encoder-decoder architectures, the estimated depth maps lack structure details and clear edges due to simple repeated upsampling. To solve this problem, this paper presents the novel LapUNet (Laplacian U-shape networks), in which the encoder adopts ResNeXt101, and the decoder is constructed with the novel DLRU (dynamic Laplacian residual U-shape) module. The DLRU module based on the U-shape structure can supplement high-frequency features by fusing dynamic Laplacian residual into the process of upsampling, and the residual is dynamically learnable due to the addition of convolutional operation. Also, the ASPP (atrous spatial pyramid pooling) module is introduced to capture image context at multiple scales though multiple parallel atrous convolutional layers, and the depth map fusion module is used for combining high and low frequency features from depth maps with different spatial resolution. Experiments demonstrate that the proposed model with moderate model size is superior to other previous competitors on the KITTI and NYU Depth V2 datasets. Furthermore, 3D reconstruction and target ranging by utilizing the estimated depth maps prove the effectiveness of our proposed method.

Full-automatic high-precision scene 3D reconstruction method with water-area intelligent complementation and mesh optimization for UAV images

ABSTRACT Fast and high-precision urban scene 3D modeling is the foundational data infrastructure for the digital earth and smart cities. However, due to challenges such as water-area matching difficulties and issues like data redundancy and insufficient observations, existing full-automatic 3D modeling methods often result in water-area missing and many small holes in the models and insufficient local-model accuracy. To overcome these challenges, full-automatic high-precision scene 3D reconstruction method with water-area intelligent complementation on depth maps and mesh optimization is proposed. Firstly, SfM was used to calculated image poses and PatchMatch was used to generated initial depth maps. Secondly, a simplified GAN extracted water-area masks and ray tracing was used achieve high-precision auto-completed water-area depth values. Thirdly, fully connected CRF optimized water-areas and arounds in depth maps. Fourthly, high-precision 3D point clouds were obtained using depth map fusion based on clustering culling and depth least squares. Then, mesh was generated and optimized using similarity measurement and vertex gradients to obtain refined mesh. Finally, high-precision scene 3D models without water-area missing or holes were generated. The results showed that: to compare with the-state-of-art ContextCapture, the proposed method enhances model completeness by 14.3%, raises average accuracy by 14.5% and improves processing efficiency by 63.6%.

A depth map fusion algorithm with improved efficiency considering pixel region prediction

A depth map fusion algorithm fuses depth maps from different perspectives into a unified coordinate framework and performs surface calculations to generate dense point clouds of the entire scene. The existing algorithms ensure the quality of these dense point clouds by eliminating inconsistencies between depth maps, but the problem of many redundant calculations often arises. In this paper, a depth map fusion algorithm based on pixel region prediction is proposed. First, the image combination is calculated to select a set of candidate neighbor images for each reference image in the scene. Second, voxels and measure estimates are constructed on a coarse scale, and an inference strategy and a corrector are proposed to merge pixel regions at a fine scale guided by the coarse scale. Finally, the deduced pixel regions at the fine scale are used as the image-space constraints for depth fusion. Public and actual oblique images datasets are used for experimental verification. Compared with the famous COLMAP, OPENMVS, Gipuma and ACMP methods, the number of redundant calculations is significantly reduced; according to Data1 ∼ Data9 in the experiment, as the number of images increases, the fusion efficiency is increased by 47.5% to 156.6%; at the same time, the point cloud accuracy is comparable to other methods.

Multi-Resolution Monocular Depth Map Fusion by Self-Supervised Gradient-Based Composition

Monocular depth estimation is a challenging problem on which deep neural networks have demonstrated great potential. However, depth maps predicted by existing deep models usually lack fine-grained details due to convolution operations and down-samplings in networks. We find that increasing input resolution is helpful to preserve more local details while the estimation at low resolution is more accurate globally. Therefore, we propose a novel depth map fusion module to combine the advantages of estimations with multi-resolution inputs. Instead of merging the low- and high-resolution estimations equally, we adopt the core idea of Poisson fusion, trying to implant the gradient domain of high-resolution depth into the low-resolution depth. While classic Poisson fusion requires a fusion mask as supervision, we propose a self-supervised framework based on guided image filtering. We demonstrate that this gradient-based composition performs much better at noisy immunity, compared with the state-of-the-art depth map fusion method. Our lightweight depth fusion is one-shot and runs in real-time, making it 80X faster than a state-of-the-art depth fusion method. Quantitative evaluations demonstrate that the proposed method can be integrated into many fully convolutional monocular depth estimation backbones with a significant performance boost, leading to state-of-the-art results of detail enhancement on depth maps. Codes are released at https://github.com/yuinsky/gradient-based-depth-map-fusion.

Automatic Measurement of Inclination Angle of Utility Poles Using 2D Image and 3D Point Cloud

The utility pole inclination angle is an important parameter for determining pole health conditions. Without depth information, the angle cannot be estimated from a 2D image, and without large labeled reference pole data, it is time consuming to locate the pole in the 3D point cloud. Therefore, this paper proposes a method that processes the pole data from the 2D image and 3D point cloud to automatically measure the pole inclination angle. Firstly, the mask of the pole skeleton is obtained from an improved Mask R-CNN. Secondly, the pole point cloud is extracted from a PointNet that deals with the generated frustum from the pole skeleton mask and depth map fusion. Finally, the angle is calculated by fitting the central axis of the pole cloud data. ApolloSpace open dataset and laboratory data are used for evaluation. The experimental results show that the AP75 of improved Mask R-CNN is 58.15%, the accuracy of PointNet is 92.4%, the average error of pole inclination is 0.66°, and the variance is 0.12°. It is proved that the method can effectively realize the automatic measurement of pole inclination.

Probabilistic Fusion of Depth Maps With a Reliable Estimation of the Local Reconstruction Quality

Depth sensors in general, but especially consumer level sensors, have noise characteristics that are difficult to generalize due to external factors such as material specific properties or varying lighting conditions. The quality of the reconstructed surface from fused depth data may therefore vary and some surface segments may be more accurately reconstructed than others. High average accuracy is important but it must be supported by an estimation of the local quality if subsequent algorithms, such as robotic grasping, use the surface reconstruction. We propose a novel approach which models sensor noise depending on local object properties and incrementally refines surface estimations using Bayesian updates. Experiments demonstrate that our algorithm reconstructs challenging scenes of the ROBI dataset (Yang et al., 2021) with comparatively high accuracy and significantly fewer outliers. We additionally provide a reliable, local estimation of the surface uncertainty.

Research on Multi-View 3D Reconstruction Technology Based on SFM.

Multi-view 3D reconstruction technology is used to restore a 3D model of practical value or required objects from a group of images. This paper designs and implements a set of multi-view 3D reconstruction technology, adopts the fusion method of SIFT and SURF feature-point extraction results, increases the number of feature points, adds proportional constraints to improve the robustness of feature-point matching, and uses RANSAC to eliminate false matching. In the sparse reconstruction stage, the traditional incremental SFM algorithm takes a long time, but the accuracy is high; the traditional global SFM algorithm is fast, but its accuracy is low; aiming at the disadvantages of traditional SFM algorithm, this paper proposes a hybrid SFM algorithm, which avoids the problem of the long time consumption of incremental SFM and the problem of the low precision and poor robustness of global SFM; finally, the MVS algorithm of depth-map fusion is used to complete the dense reconstruction of objects, and the related algorithms are used to complete the surface reconstruction, which makes the reconstruction model more realistic.

Accelerated Multi-View Stereo for 3D Reconstruction of Transmission Corridor with Fine-Scale Power Line

Fast reconstruction of power lines and corridors is a critical task in UAV (unmanned aerial vehicle)-based inspection of high-voltage transmission corridors. However, recent dense matching algorithms suffer the problem of low efficiency when processing large-scale high-resolution UAV images. This study proposes an efficient dense matching method for the 3D reconstruction of high-voltage transmission corridors with fine-scale power lines. First, an efficient random red-black checkerboard propagation is proposed, which utilizes the neighbor pixels with the most similar color to propagate plane parameters. To combine the pixel-wise view selection strategy adopted in Colmap with the efficient random red-black checkerboard propagation, the updating schedule for inferring visible probability is improved; second, strategies for decreasing the number of matching cost computations are proposed, which can reduce the unnecessary hypotheses for verification. The number of neighbor pixels necessary to propagate plane parameters is reduced with the increase of iterations, and the number of the combinations of depth and normal is reduced for the pixel with better matching cost in the plane refinement step; third, an efficient GPU (graphics processing unit)-based depth map fusion method is proposed, which employs a weight function based on the reprojection errors to fuse the depth map. Finally, experiments are conducted by using three UAV datasets, and the results indicate that the proposed method can maintain the completeness of power line reconstruction with high efficiency when compared to other PatchMatch-based methods. In addition, two benchmark datasets are used to verify that the proposed method can achieve a better F1 score, 4–7 times faster than Colmap.

Deep Binocular Fixation Prediction Using a Hierarchical Multimodal Fusion Network

RGB-D data are increasingly being used for myriad computer vision tasks. For such tasks, most methods simply concatenate or add feature vectors from RGB images and depth maps and allow the two modalities to complement each other mutually. However, such a fusion strategy results in inefficient and inadequate performance. In this study, we propose deep binocular fixation prediction based on a hierarchical multimodal fusion network that suitably combines RGB and depth maps. In the proposed method, a novel convolutional block attention module completely extracts image texture features and retains spatial information. In addition, a pyramid dilated-convolution module refines feature information, further improving the fusion of RGB and depth maps. Experimental results indicate that the proposed network achieves state-of-the-art performance on the NUS and NCTU datasets.

MULTI-VIEW STEREO WITH SEMANTIC PRIORS

Abstract. Patch-based stereo is nowadays a commonly used image-based technique for dense 3D reconstruction in large scale multi-view applications. The typical steps of such a pipeline can be summarized in stereo pair selection, depth map computation, depth map refinement and, finally, fusion in order to generate a complete and accurate representation of the scene in 3D. In this study, we aim to support the standard dense 3D reconstruction of scenes as implemented in the open source library OpenMVS by using semantic priors. To this end, during the depth map fusion step, along with the depth consistency check between depth maps of neighbouring views referring to the same part of the 3D scene, we impose extra semantic constraints in order to remove possible errors and selectively obtain segmented point clouds per label, boosting automation towards this direction. In order to reassure semantic coherence between neighbouring views, additional semantic criterions can be considered, aiming to eliminate mismatches of pixels belonging in different classes.

Prioritized multi-view stereo depth map generation using confidence prediction

In this work, we propose a novel approach to prioritize the depth map computation of multi-view stereo (MVS) to obtain compact 3D point clouds of high quality and completeness at low computational cost. Our prioritization approach operates before the MVS algorithm is executed and consists of two steps. In the first step, we aim to find a good set of matching partners for each view. In the second step, we rank the resulting view clusters (i.e. key views with matching partners) according to their impact on the fulfillment of desired quality parameters such as completeness, ground resolution and accuracy. Additional to geometric analysis, we use a novel machine learning technique for training a confidence predictor. The purpose of this confidence predictor is to estimate the chances of a successful depth reconstruction for each pixel in each image for one specific MVS algorithm based on the RGB images and the image constellation. The underlying machine learning technique does not require any ground truth or manually labeled data for training, but instead adapts ideas from depth map fusion for providing a supervision signal. The trained confidence predictor allows us to evaluate the quality of image constellations and their potential impact to the resulting 3D reconstruction and thus builds a solid foundation for our prioritization approach. In our experiments, we are thus able to reach more than 70% of the maximal reachable quality fulfillment using only 5% of the available images as key views. For evaluating our approach within and across different domains, we use two completely different scenarios, i.e. cultural heritage preservation and reconstruction of single family houses.

Rapid 3D Reconstruction for Image Sequence Acquired from UAV Camera.

In order to reconstruct three-dimensional (3D) structures from an image sequence captured by unmanned aerial vehicles’ camera (UAVs) and improve the processing speed, we propose a rapid 3D reconstruction method that is based on an image queue, considering the continuity and relevance of UAV camera images. The proposed approach first compresses the feature points of each image into three principal component points by using the principal component analysis method. In order to select the key images suitable for 3D reconstruction, the principal component points are used to estimate the interrelationships between images. Second, these key images are inserted into a fixed-length image queue. The positions and orientations of the images are calculated, and the 3D coordinates of the feature points are estimated using weighted bundle adjustment. With this structural information, the depth maps of these images can be calculated. Next, we update the image queue by deleting some of the old images and inserting some new images into the queue, and a structural calculation of all the images can be performed by repeating the previous steps. Finally, a dense 3D point cloud can be obtained using the depth–map fusion method. The experimental results indicate that when the texture of the images is complex and the number of images exceeds 100, the proposed method can improve the calculation speed by more than a factor of four with almost no loss of precision. Furthermore, as the number of images increases, the improvement in the calculation speed will become more noticeable.

On-Board Object Detection: Multicue, Multimodal, and Multiview Random Forest of Local Experts.

Despite recent significant advances, object detection continues to be an extremely challenging problem in real scenarios. In order to develop a detector that successfully operates under these conditions, it becomes critical to leverage upon multiple cues, multiple imaging modalities, and a strong multiview (MV) classifier that accounts for different object views and poses. In this paper, we provide an extensive evaluation that gives insight into how each of these aspects (multicue, multimodality, and strong MV classifier) affect accuracy both individually and when integrated together. In the multimodality component, we explore the fusion of RGB and depth maps obtained by high-definition light detection and ranging, a type of modality that is starting to receive increasing attention. As our analysis reveals, although all the aforementioned aspects significantly help in improving the accuracy, the fusion of visible spectrum and depth information allows to boost the accuracy by a much larger margin. The resulting detector not only ranks among the top best performers in the challenging KITTI benchmark, but it is built upon very simple blocks that are easy to implement and computationally efficient.

Geometry Guided Multi-Scale Depth Map Fusion via Graph Optimization.

In depth discontinuous and untextured regions, depth maps created by multiple view stereopsis are with heavy noises, but existing depth map fusion methods cannot handle it explicitly. To tackle the problem, two novel strategies are proposed: 1) a more discriminative fusion method, which is based on geometry consistency, measuring the consistency, and stability of surface geometry computed on both partial and global surfaces, different from traditional methods only using visibility consistency; 2) a graph optimization method which fuses pyramids of depth maps as mutual complementary information is available in different scales, and differs from existing multi-scale fusion methods. The method considers both sampling scale of a point and relations among points, and is proven to be solvable by graph cuts. Experimental results verify the superior performance of the proposed method to the traditional visibility consistency-based methods, and the proposed method is also compared favorably with a number of state-of-the-art methods. Moreover, the proposed method achieves the highest completeness among all the methods compared.

Improved marching tetrahedra algorithm based on hierarchical signed distance field and multi-scale depth map fusion for 3D reconstruction

3D reconstruction systems are promoted by developments of both computer hardware and computing technologies. They still remain problems like high expense, low efficiency and inaccuracy. Especially for large-scale scenes, lack of full use of multi-scale depth information will cause blurring and irreal reconstruction results. To solve this problem, we construct the structure of hierarchical signed distance field (H-SDF) and design an improved marching tetrahedra algorithm for multi-scale depth map fusion. In addition, to improve efficiency, we also propose a two-phase search strategy in image feature matching: the bag-of-features model (BOF) is adopted in a coarse search to narrow search scope and then the SIFT descriptor is used in exact matching to pick reconstruction image points. Experiment results indicate that coarse search makes matching time shorter; using the H-SDF to fuse multi-scale depth maps, and isosurface extraction with improved marching tetrahedra algorithm can improve visual effect.

A MEDIAN-BASED DEPTHMAP FUSION STRATEGY FOR THE GENERATION OF ORIENTED POINTS

Abstract. Due to good scalability, systems for image-based dense surface reconstruction often employ stereo or multi-baseline stereo methods. These types of algorithms represent the scene by a set of depth or disparity maps which eventually have to be fused to extract a consistent, non-redundant surface representation. Generally the single depth observations across the maps possess variances in quality. Within the fusion process not only preservation of precision and detail but also density and robustness with respect to outliers are desirable. Being prune to outliers, in this article we propose a local median-based algorithm for the fusion of depth maps eventually representing the scene as a set of oriented points. Paying respect to scalability, points induced by each of the available depth maps are streamed to cubic tiles which then can be filtered in parallel. Arguing that the triangulation uncertainty is larger in the direction of image rays we define these rays as the main filter direction. Within an additional strategy we define the surface normals as the principle direction for median filtering/integration. The presented approach is straight-forward to implement since employing standard oc- and kd-tree structures enhanced by nearest neighbor queries optimized for cylindrical neighborhoods. We show that the presented method in combination with the MVS (Rothermel et al., 2012) produces surfaces comparable to the results of the Middlebury MVS benchmark and favorably compares to an state-of-the-art algorithm employing the Fountain dataset (Strecha et al., 2008). Moreover, we demonstrate its capability of depth map fusion for city scale reconstructions derived from large frame airborne imagery.

A MEDIAN-BASED DEPTHMAP FUSION STRATEGY FOR THE GENERATION OF ORIENTED POINTS

Due to good scalability, systems for image-based dense surface reconstruction often employ stereo or multi-baseline stereo methods. These types of algorithms represent the scene by a set of depth or disparity maps which eventually have to be fused to extract a consistent, non-redundant surface representation. Generally the single depth observations across the maps possess variances in quality. Within the fusion process not only preservation of precision and detail but also density and robustness with respect to outliers are desirable. Being prune to outliers, in this article we propose a local median-based algorithm for the fusion of depth maps eventually representing the scene as a set of oriented points. Paying respect to scalability, points induced by each of the available depth maps are streamed to cubic tiles which then can be filtered in parallel. Arguing that the triangulation uncertainty is larger in the direction of image rays we define these rays as the main filter direction. Within an additional strategy we define the surface normals as the principle direction for median filtering/integration. The presented approach is straight-forward to implement since employing standard oc- and kd-tree structures enhanced by nearest neighbor queries optimized for cylindrical neighborhoods. We show that the presented method in combination with the MVS (Rothermel et al., 2012) produces surfaces comparable to the results of the Middlebury MVS benchmark and favorably compares to an state-of-the-art algorithm employing the Fountain dataset (Strecha et al., 2008). Moreover, we demonstrate its capability of depth map fusion for city scale reconstructions derived from large frame airborne imagery.

Data-driven contextual modeling for 3D scene understanding

The recent development of fast depth map fusion technique enables the realtime, detailed scene reconstruction using commodity depth camera, making the indoor scene understanding more possible than ever. To address the specific challenges in object analysis at subscene level, this work proposes a data-driven approach to modeling contextual information covering both intra-object part relations and inter-object object layouts. Our method combines the detection of individual objects and object groups within the same framework, enabling contextual analysis without knowing the objects in the scene a priori. The key idea is that while contextual information could benefit the detection of either individual objects or object groups, both can contribute to object extraction when objects are unknown.Our method starts with a robust segmentation and partitions a subscene into segments, each of which represents either an independent object or a part of some object. A set of classifiers are trained for both individual objects and object groups, using a database of 3D scene models. We employ the multiple kernel learning (MKL) to learn per-category optimized classifiers for objects and object groups. Finally, we perform a graph matching to extract objects using the classifiers, thus grouping the segments into either an object or an object group. The output is an object-level labeled segmentation of the input subscene. Experiments demonstrate that the unified contextual analysis framework achieves robust object detection and recognition over cluttered subscenes.

A Unified Probabilistic Framework for Real-Time Depth Map Fusion

This paper proposes a unified probabilistic framework for real-time depth map fusion. By modeling the depth imaging process as a random experiment, the depth map fusion is converted into probability density function (pdf) estimation. The depth fusion problem is decoupled into four parts: the fusion space, the influence term, the visibility term and the confidence term. We combine these four terms in a unified probabilistic framework, and apply the framework in two cases to evaluate the performance. In the first case, multiple stereo vision cameras are used to acquire multiple depth map streams from multiple viewpoints simultaneously in real time. In the second case, two cameras and a Kinect are combined to provide two depth map streams. In both cases, strategies for each part of the framework are presented to perform real-time depth fusion. Experimental results show that the proposed framework is promising for real-time depth map fusion.

Real-time 3D reconstruction at scale using voxel hashing

Online 3D reconstruction is gaining newfound interest due to the availability of real-time consumer depth cameras. The basic problem takes live overlapping depth maps as input and incrementally fuses these into a single 3D model. This is challenging particularly when real-time performance is desired without trading quality or scale. We contribute an online system for large and fine scale volumetric reconstruction based on a memory and speed efficient data structure. Our system uses a simple spatial hashing scheme that compresses space, and allows for real-time access and updates of implicit surface data, without the need for a regular or hierarchical grid data structure. Surface data is only stored densely where measurements are observed. Additionally, data can be streamed efficiently in or out of the hash table, allowing for further scalability during sensor motion. We show interactive reconstructions of a variety of scenes, reconstructing both fine-grained details and large scale environments. We illustrate how all parts of our pipeline from depth map pre-processing, camera pose estimation, depth map fusion, and surface rendering are performed at real-time rates on commodity graphics hardware. We conclude with a comparison to current state-of-the-art online systems, illustrating improved performance and reconstruction quality.