論文

© 2019, International Sports Engineering Association. This paper presents a method to estimate a time-sequential trajectory of the center of mass (CoM) of an athlete from a multi-view set of cameras. Collecting the CoM typically requires large-scale measuring systems or attaching sensors to the athletes. To mitigate such hardware limitations, the present study takes a multi-view video-based approach. The proposed method reconstructs subjects’ voxels from a set of multi-view frames and weights each voxel with body part-dependent weights to calculate a CoM. Our results, using real data measured in a studio, showed that the proposed method can estimate CoM within 20 mm concerning center of pressure measures.

© 2020, Springer Nature Singapore Pte Ltd. Estimating accurate depth from an RGB image in any environment is challenging task in computer vision. Recent learning based method using deep Convolutional Neural Networks (CNNs) have driven plausible appearance, but these conventional methods are not good at estimating scenes that have a pure rotation of camera, such as in-plane rolling. This movement imposes perturbations on learning-based methods because gravity direction is considered to be strong prior to CNN depth estimation (i.e., the top region of an image has a relatively large depth, whereas bottom region tends to have a small depth). To overcome this crucial weakness in depth estimation with CNN, we propose a simple but effective refining method that incorporates in-plane roll alignment using camera poses of monocular Simultaneous Localization and Mapping (SLAM). For the experiment, we used public datasets and also created our own dataset composed of mostly in-plane roll camera movements. Evaluation results on these datasets show the effectiveness of our approach.

© 2019 IEEE. This work addresses the task of open world semantic segmentation using RGBD sensing to discover new semantic classes over time. Although there are many types of objects in the real-word, current semantic segmentation methods make a closed world assumption and are trained only to segment a limited number of object classes. Towards a more open world approach, we propose a novel method that incrementally learns new classes for image segmentation. The proposed system first segments each RGBD frame using both color and geometric information, and then aggregates that information to build a single segmented dense 3D map of the environment. The segmented 3D map representation is a key component of our approach as it is used to discover new object classes by identifying coherent regions in the 3D map that have no semantic label. The use of coherent region in the 3D map as a primitive element, rather than traditional elements such as surfels or voxels, also significantly reduces the computational complexity and memory use of our method. It thus leads to semi-real-time performance at 10.7 Hz when incrementally updating the dense 3D map at every frame. Through experiments on the NYUDv2 dataset, we demonstrate that the proposed method is able to correctly cluster objects of both known and unseen classes. We also show the quantitative comparison with the state-of-the-art supervised methods, the processing time of each step, and the influences of each component.

© 1995-2012 IEEE. State-of-the-art methods for diminished reality propagate pixel information from a keyframe to subsequent frames for real-time inpainting. However, these approaches produce artifacts, if the scene geometry is not sufficiently planar. In this article, we present InpaintFusion, a new real-time method that extends inpainting to non-planar scenes by considering both color and depth information in the inpainting process. We use an RGB-D sensor for simultaneous localization and mapping, in order to both track the camera and obtain a surfel map in addition to RGB images. We use the RGB-D information in a cost function for both the color and the geometric appearance to derive a global optimization for simultaneous inpainting of color and depth. The inpainted depth is merged in a global map by depth fusion. For the final rendering, we project the map model into image space, where we can use it for effects such as relighting and stereo rendering of otherwise hidden structures. We demonstrate the capabilities of our method by comparing it to inpainting results with methods using planar geometric proxies.

© 2019 IEEE. Image inpainting aims to restore texture of missing regions in scene from an RGB image. In this paper, we aim to restore not only the texture but also the geometry of the missing regions in scene from a pair of RGB and depth images. Inspired by the recent development of generative adversarial network, we employ an encoder-decoderbased generative adversarial network with the input of RGB and depth image. The experimental results show that our method restores the missing region of both RGB and depth image.

© Springer Nature Switzerland AG 2019. Recent semantic segmentation systems have achieved significant improvement by performing pixel-wise training with hierarchical features using deep convolutional neural network models. While the learning process usually requires pixel-level annotated images, it is difficult to get desirable amounts of fine-labeled data and thus the training set size is more likely to be limited, often in thousands. This means that top methods for a dataset can be fine-tuned for a specific situation, making the generalization ability unclear. In real-world applications like self-driving systems, ambiguous region or lack of context information can cause errors in the predicted results. Resolving such ambiguities is crucial for subsequent operations to be performed safely. We are inspired by work from CodeSLAM where optimizable pixel-wise depth representation is learned. We modify the regression method to work on the pixel-wise classification problem. By training a variational auto-encoder network conditioned with a color image, the computed latent space works as a low-dimensional representation of semantic segmentation, which can be efficiently optimized. As a consequence, our model can correct the error or ambiguity of the prediction during the inference phase given useful scene information. We show how this approach works by giving partial scene truth and perform optimization on the latent variable.

© Springer Nature Switzerland AG 2019. In this paper, we propose a line-based SLAM from an image sequence captured by a camera mounted on a vehicle in consideration with the prior distribution of line features that detected in an urban environments. Since such scenes captured by the vehicle in urban envirounments can be expected to include a lot of line segments detected from road markings and buildings, we employ line segments as features for our SLAM. We use additional prior regarding the line segments so that we can improve the accuracy of the SLAM. We assume that the angle of the vector of the line segments to the vehicle’s direction of travel conform to four-component Gaussian mixture distribution. We define a new cost function considering the prior distribution and optimize the relative camera pose, position, and the 3D line segments by bundle adjustment. The prior distribution is also extended into 2D, the distance and angle of the line segments. In addition, we make digital maps from the detected line segments. Our method increases the accuracy of localization and corrects tilted lines in the digital maps. We apply our method to both the single-camera system and the multi-camera system for demonstrate the accuracy improvement by the prior distribution of distance and angle of line features.

© 2018 The Institute of Electronics, Information and Communication Engineers. We propose a method of line-based Simultaneous Localization and Mapping (SLAM) using non-overlapping multiple cameras for vehicles running in an urban environment. It uses corresponding line segments between images taken by different frames and different cameras. The contribution is a novel line segment matching algorithm by warping processing based on urban structures. This idea significantly improves the accuracy of line segment matching when viewing direction are very different, so that a number of correspondences between front-view and rear-view cameras can be found and the accuracy of SLAM can be improved. Additionally, to enhance the accuracy of SLAM we apply a geometrical constraint of urban area for initial estimation of 3D mapping of line segments and optimization by bundle adjustment. We can further improve the accuracy of SLAM by combining points and lines. The position error is stable within 1.5m for the entire image dataset evaluated in this paper. The estimation accuracy of our method is as high as that of ground truth captured by RTK-GPS. Our high accuracy SLAM algorithm can be apply for generating a road map represented by line segments. According to an evaluation of our generating map, true positive rate around the vehicle exceeding 70% is achieved.

© 2019 Copyright is held by the owner/author(s). ACM ISBN 978-1-4503-6889-6/19/10. The second ACM International Workshop on Multimedia Content Analysis in Sports (ACM MMSports'19) is held in Nice, France on October 25th, 2019 co-located with the ACM International Conference on Multimedia 2019 (ACM Multimedia 2019). The goal of this workshop is to bring together researchers and practitioners from academia and industry to address challenges and report progress in mining, analyzing, understanding and visualizing the multimedia/multimodal data in sports. The combination of sports and modern technology offers a novel and intriguing field of research with promising approaches for visual broadcast augmentation, understanding, statistical analysis and evaluation, and sensor fusion. There is a lack of research communities focusing on the fusion of multiple modalities. We are helping to close this research gap with this workshop series on multimedia content analysis in sports.

© 2019 IEEE. The standard photometric stereo is a technique to densely reconstruct objects' surfaces using light variation under the assumption of a static camera with a moving light source. In this work, we use photometric stereo to reconstruct dense 3D scenes while moving the camera and the light altogether. In such non-static case, camera poses as well as correspondences between pixels of each frame to apply photometric stereo are required. ORB-SLAM is a technique that can be used to estimate camera poses. To retrieve correspondences, our idea is to start from a sparse 3D mesh obtained with ORB SLAM and then densify the mesh by a plane sweep method using a multi-view photometric consistency. By combining ORB-SLAM and photometric stereo, it is possible to reconstruct dense 3D scenes with a off-the-shelf smartphone and its embedded torchlight. Note that SLAM systems usually struggle with textureless object, which is effectively compensated by the photometric stereo in our method. Experiments are conducted to show that our proposed method gives better results than SLAM alone or COLMAP, especially for partially textureless surfaces.

© 2018, Society for Imaging Science and Technology. In this paper, we introduce a method to visually inspect in- dustrial parts by using multi-view images of an industrial part on a rotating table. During the visual inspection of the parts' sur- faces, the relationship between the camera pose, the light vectors, and the normal vectors of a part's surface is the key factor in de- tecting abnormalities. We can change the relationship between the three factors by rotating the part; then, the abnormalities are visible in certain positional relationships. We, therefore, track the points on a part's surfaces in an image sequence and discrimi- nate abnormal points (such as points on scratches or dents) from normal points based on pixel value transitions while rotating the part. The experimental results, which were based on real data of industrial parts, showed that the proposed method could detect the abnormalities on the surfaces.

© 2020 by the authors. This paper presents a method for estimating the six Degrees of Freedom (6DoF) pose of texture-less primitive-shaped objects from depth images. As the conventional methods for object pose estimation require rich texture or geometric features to the target objects, these methods are not suitable for texture-less and geometrically simple shaped objects. In order to estimate the pose of the primitive-shaped object, the parameters that represent primitive shapes are estimated. However, these methods explicitly limit the number of types of primitive shapes that can be estimated. We employ superquadrics as a primitive shape representation that can represent various types of primitive shapes with only a few parameters. In order to estimate the superquadric parameters of primitive-shaped objects, the point cloud of the object must be segmented from a depth image. It is known that the parameter estimation is sensitive to outliers, which are caused by the miss-segmentation of the depth image. Therefore, we propose a novel estimation method for superquadric parameters that are robust to outliers. In the experiment, we constructed a dataset in which the person grasps and moves the primitive-shaped objects. The experimental results show that our estimation method outperformed three conventional methods and the baseline method.

© 2020 IEEE. This paper addresses the task of six degrees of freedom (6-DoF) pose estimation of stacked rectangular objects from depth images. Object pose estimation is one of the key challenges for visual processing systems since it plays a vital role in many situations such as warehouse/factory automation, robotic manipulation, and augmented reality. Many recent approaches to object pose estimation use RGB information for detecting and estimating the pose of objects. However, in warehouse/factory automation, objects are often small, occluded, cluttered, and texture-less which makes it difficult to utilize RGB features for detection and pose estimation. In order to overcome this restriction, we use only the depth information (without RGB information) and its geometric features to segment each object and to estimate the 6-DoF (position and orientation) in a stacked scene. We segment the rectangular objects in each scene from the depth and surface normal discontinuities (geometric segmentation). From the geometrically segmented image, four object corner points can be estimated using the convex hull detection and the eight corner points, which are required for the 6-DoF pose estimation, can be calculated. To improve the accuracy of orientation estimation, we estimate four orientation candidates and select the best among them. Experimental results using two evaluation methods show that our method outperformed the baseline method.

© 2019 Copyright held by the owner/author(s). In this paper, we propose a method to predict the future shot direction in a tennis match using pose information and player position. As far as we know, there is no work that deals with such a predictive task, so there is no shot direction dataset as yet. Therefore, using a YouTube tennis match video, we construct an time of impact and shot direction dataset. To reduce annotation costs, we propose a method to automatically label the shot direction. Moreover, we propose a method to predict the future shot direction using the constructed dataset. The shot direction is predicted using LSTM(long short-time memory), from sequential pose information up to the time of impact and the player position. We employ OpenPose to extract the position of skeleton joints. In the experiment, we evaluate the accuracy of shot direction prediction and verify the effectiveness of the proposed method. Since there are no studies that predict future shot direction, we set four baseline methods to evaluate the effectiveness of our proposed method.

© 2018, The Author(s). Diminishing the appearance of a fence in an image is a challenging research area due to the characteristics of fences (thinness, lack of texture, etc.) and the need for occluded background restoration. In this paper, we describe a fence removal method for an image sequence captured by a user making a sweep motion, in which occluded background is potentially observed. To make use of geometric and appearance information such as consecutive images, we use two well-known approaches: structure from motion and light field rendering. Results using real image sequences show that our method can stably segment fences and preserve background details for various fence and background combinations. A new video without the fence, with frame coherence, can be successfully provided.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. Human motion capture (MoCap) plays a key role in healthcare and human–robot collaboration. Some researchers have combined orientation measurements from inertial measurement units (IMUs) and positional inference from cameras to reconstruct the 3D human motion. Their works utilize multiple cameras or depth sensors to localize the human in three dimensions. Such multiple cameras are not always available in our daily life, but just a single camera attached in a smart IP devices has recently been popular. Therefore, we present a 3D pose estimation approach from IMUs and a single camera. In order to resolve the depth ambiguity of the single camera configuration and localize the global position of the subject, we present a constraint which optimizes the foot-ground contact points. The timing and 3D positions of the ground contact are calculated from the acceleration of IMUs on foot and geometric transformation of foot position detected on image, respectively. Since the results of pose estimation is greatly affected by the failure of the detection, we design the image-based constraints to handle the outliers of positional estimates. We evaluated the performance of our approach on public 3D human pose dataset. The experiments demonstrated that the proposed constraints contributed to improve the accuracy of pose estimation in single and multiple camera setting.

© Springer Nature Switzerland AG 2020. Diminished reality is a technology that aims to remove objects from video images and fills in the missing region with plausible pixels. Most conventional methods utilize the different cameras that capture the same scene from different viewpoints to allow regions to be removed and restored. In this paper, we propose an RGB-D image inpainting method using generative adversarial network, which does not require multiple cameras. Recently, an RGB image inpainting method has achieved outstanding results by employing a generative adversarial network. However, RGB inpainting methods aim to restore only the texture of the missing region and, therefore, does not recover geometric information (i.e, 3D structure of the scene). We expand conventional image inpainting method to RGB-D image inpainting to jointly restore the texture and geometry of missing regions from a pair of RGB and depth images. Inspired by other tasks that use RGB and depth images (e.g., semantic segmentation and object detection), we propose late fusion approach that exploits the advantage of RGB and depth information each other. The experimental results verify the effectiveness of our proposed method.

© 2019 IEEE. The measurement of range of motion during medical examination is a basic procedure in diagnosis and treatment evaluation. However, accurate measurement requires experience, and it is difficult for patients to measure range of motion by themselves. If patients can easily and accurately measure the range by their own, it will encourage patients to continue rehabilitation and will be useful for helping patients to determine when they should see a doctor. We developed a wrist flexion and extension measurement system using images captured by a smart phone so that patients can measure range of motion without expert knowledge or experience. First, the system calculates a middle line for the forearm and an outside line for the dorsal hand based on user inputs. Afterward, the angle formed by these lines is calculated. The experimental results show that the average absolute errors are 4.82° for flexion and 8.74° for extension.

© 2019 IEEE. This paper presents a method for the virtual manipulation of real living space using semantic segmentation of a 3D point cloud captured in the real world. We applied PointNet to segment each piece of furniture from the point cloud of a real indoor environment captured by moving a RGB-D camera. For semantic segmentation, we focused on local geometric information not used in PointNet, and we proposed a method to refine the class probability of labels attached to each point in PointNet's output. The effectiveness of our method was experimentally confirmed. We then created 3D models of real-world furniture using a point cloud with corrected labels, and we virtually manipulated real living space using Dollhouse VR, a layout system.

© Springer Nature Switzerland AG 2019. This paper presents a pipeline for stereo visual odometry using cameras with different fields of view. It gives a proof of concept about how a constraint on the respective field of view of each camera can lead to both an accurate 3D reconstruction and a robust pose estimation. Indeed, when considering a fixed resolution, a narrow field of view has a higher angular resolution and can preserve image texture details. On the other hand, a wide field of view allows to track features over longer periods since the overlap between two successive frames is more substantial. We propose a semi-independent stereo system where each camera performs individually temporal multi-view optimization but their initial parameters are still jointly optimized in an iterative framework. Furthermore, the concept of lead and follow camera is introduced to adaptively propagate information between the cameras. We evaluate the method qualitatively on two indoor datasets, and quantitatively on a synthetic dataset to allow the comparison across different fields of view.