In June 2016, all local stations of NHK began operating File-Based News Production System. In this system, all users are able to share video files via IP network and use meta data of contents effectively. This file-based system realizes dramatic improvements in speed and efficiency for the workflow of news production.

抄録全体を表示

In this paper, we propose a new geometric verification method with an epipolar constraint. The proposed method reduces the number of inliers for incorrect image pairs by filtering geometrically inconsistent local feature correspondences and detecting invalid camera poses from estimated epipolar geometry during RANSAC iterations.

抄録全体を表示

A high - speed gaze controller enables continuous high - resolution imaging of dynamic objects, and introducing a wide - angle camera can improve continuity of high - speed tracking. We introduce a fractal calibration pattern against their greatly different focal length s , and achieve a calibrated system of the controller and the camera efficiently.

抄録全体を表示

We have developed an accurate calibration method for an active multi-view camera system using angle sensors. In this paper, we propose a technique that deals with various camera placements in filming locations and improves the accuracy of the calibration. The effectiveness of the proposed technique demonstrated through a experiment.

抄録全体を表示

In this paper, we propose a method to estimate object size from a single image taken with monocular camera. We aim at the method that can ea sily estimate size only by giving the size of a known object without using range sensor s and stereo camera s .

抄録全体を表示

In this paper, we propose a method to realize volunteer computing using Hybridcast. We focus on idle computing resources of TV sets which viewers watch broadcast program on. We developed a prototype system and conducted an experiment. The experimental results show the effectiveness of our method, and reveal some problems particular to Hybrid cast browser.

抄録全体を表示

We constructed a model of receiving 4K Live video by TV over the Hybridcast system,and we report our experimented model.4K-Live video is encoded HEVC,and segmented Mpeg-Dash.encoded file is uploaded to Web server. Live video is played at realtime on the TV receiver by Mpeg-Dash Player over the Hybridcast system.

抄録全体を表示

4K/8K systems are expected to take over HDTV. Commercial 4K broadcasting has started and the 8K broadcasting is planned for 2018. However, the content is still insufficient. It is a similar situation to that of HDTV in the 1990s’. HDTV content upconverted from analogue TV content was important to supplement the insufficient HDTV content. The upconverted HDTV content was also important for news coverage because HDTV equipment was heavy and bulky. Current 4K/8K situation is similar and covering the news with 8K equipment is very difficult because 8K equipment is much heavier and bulkier than that of HDTV in 1990s’. Currently HDTV content is sufficient and HDTV equipment has matured to cope with news coverage. Upconverter from HDTV to 8K is a variable candidate to solve the problem. However the upconversion from the interlaced HDTV to 8K is 32 times enlargement and the upconverted 8K images become very blurry. The upconverter from HDTV to 8K with super resolution is proposed to fix the blur. In this paper the details are discussed.

抄録全体を表示

The 8K video is expected w ide utilization f rom broadcast to medical treatment and d igital s ignage. However, to make 8K video is difficult for loca l station. Therefore , we introduce case studies of 8K production in local stations.

抄録全体を表示

In this article, an 8-tap CMOS lock-in pixel with lateral electric field charge modulator is presented and its characteristics are evaluated by device simulations. The proposed pixel makes possible to measure the highly time-resolved images with a high signal to noise ratio and to observe the various imaging cells having different fluorescence lifetimes in real time.

抄録全体を表示

We have developed a polarization - analyzing CMOS image sensor. We made a palm - size in - line chiral analysis device for microchemistry systems, and succeeded in real - time measurements. In this work, we designed a new polarization - analyzing device with improved performance.

抄録全体を表示

We have developed a light pipe array and constructed a miniaturized digital ELISA system which consisted of the light pipe array, a PDMS microfluidic and a CMOS image sensor. In this work , we evaluated sensitivity of the digital ELISA system

抄録全体を表示

We propose the method of light-adaptive imaging. Using interleave-integration-control image sensor, we resolve the problem that trade-off of the motion blur and S/N in conventional SVE systems. The ratio of sensitivity is variable by controlling the exposure time pattern in a 2×2 pixel block unit. In this paper, we explain how to reconstruct a high quality image, and show effectiveness of the proposed method by a simulation.

抄録全体を表示

This paper proposes an effective retrieval of inspection records in expressway maintenance has been proposed. A realization of support inspection system is a very important task in the sustainable infrastructure. In the proposed method, experienced inspectors' evaluation of the previous retrieval result is introduced to the our retrieval system.

抄録全体を表示

Apparent motion estimation and obstructive objects removal technique on a video are closely related and important techniques for video editing. We consider a basic removal technique based on the motion estimation under obstructive objects and evaluate its basic performance.

抄録全体を表示

In this paper we applied the Caffe convolutional neural networ k (CNN) framework to portrait aesthetics quality estimation. We d esign a set of semantic complexity features extracted from th e intermediate layer features of the CNN architecture. Besides , CNN features trained for different task are compared in aesthe t ic s estimation, and the contribution from content is identifie d .

抄録全体を表示

An image retrieval system based on name of person/place which reflected in news video images is required to utilize archived videos. One of the techniques to realize the system is a scene text recognition technique which can recognize a text written on signs and name tags. We propose a robust method of detecting multi-oriented scene text candidate by using a rotated-aspect ratio feature. Moreover, we define a rotated bounding box descriptor to fast find the rotated-aspect ratio.

抄録全体を表示

PatchMatch Multi-View Stereo (MVS) is a method generating a depth map from multi-view images. The drawback of PatchMatch MVS is its large computational amount. We presents an FPGA accelerator for PatchMatch MVS for high-speed and low-power; this is designed using OpenCL that allows users a high-level description like C/C++ for efficient design.

抄録全体を表示

We propose a method for Alzheimer's Disease classification without anatomical regions of interest. We perform a majority voting weighted by a prior distribution that is constructed based on a Boosting algorithm. Through the experiments, we demonstrate the effectiveness of the proposed method using Alzheimer's Disease Neuroimaging Initiative dataset.

抄録全体を表示

Augmented Reality (AR) is one of the most interesting and useful technologies. A variety applications using AR are expected. We tried to study a coloring tool for kids as one of the application. In the application, CG fish with texturing the handwriting picture is superimposed on the video fromWEB camera.

抄録全体を表示

We present an novel inter-person occlusion handling method for robust multi-person tracking-by-detection using a single omnidirectional camera. The proposed method matches particle filter trackers with detection areas based on the two types of matching scores: one-to-one matching score for single person tracking and many-to-one matching score for occlusion handling. The experimental results show that the proposed method achieves robust multi-person tracking and resolves 63% of the tracking failures of the conventional method for a complex video sequence of 11 people walking around randomly.

抄録全体を表示

We have developed a color-grading equipment for full-featured 8K Super Hi-Vision program production. This equipment corresponds to the ultrahigh-definition signal/data interface (U-SDI) and can perform real-time color correction, flaw correction, chromatic aberration correction, and detail correction. In addition, down-conversion and crop-out functions are installed for detailed checks using an HD or a 4K monitor. In order to realize these functions, two FPGAs are implemented. In this report, we discuss the detailed configuration and performance of this equipment.

抄録全体を表示

We have developed a seamless switcher for switching full-featured 8K video signals (approximately 144Gbps) with Ultrahigh-definition Signal/Data Interfaces (U-SDI). Considering the extensibility for an increased number of channels, we used cross point switch devices rather than FPGAs.

抄録全体を表示

We have a development of 8K/120fps. Synchronization in 30fps systems, tri-level sync and black-burst (B.B.) signal have been used. On the other hand, the standard was formulated for generating a video sync signal from the precision time synchronization by PTP. Through verification experiment by prototype 8K/120fps devices, we studied synchronizing signal suitable for 120fps video production.

抄録全体を表示

This report discusses two methods of measuring MTF for UHDTV cameras using a ladder chart and slanted edge.

抄録全体を表示

We have developed an 8K “full-resolution”, 60 Hz, portable camera system using a 133-megapixel CMOS single-chip image sensor. The weight of the camera head is 6.3 kg and the size of the CCU is 3U. Measurement results show limiting resolution is over 4000 TV Lines and signal-to-noise ratio is 51 dB at the sensitivity of F5.6 at 2000 lux, dynamic range of 200%.

抄録全体を表示

In X - r ay detector used for CT, material identification is possible with Photon counting image sensor , it is not possible in widely used traditional image sensor . To do material identification using traditional image sensor , signal detection method depends on the energy of X - rays using the physical inte raction of X - rays was examined.

抄録全体を表示

The purpose of this study is to show superiority of photon - counting when radiation dose is low. In this paper, we captured data by photon - counting and conventional method, and evaluate d degradation by difference method. In the result, degradation of photon - counting is less, and we confirmed superiority of photon - counting.

抄録全体を表示

In this study, we propose a spectral computed tomography (CT) with reduced data tra ffic. Energy spectra taken from selected pixels were combined with conventional CT image to allow material identification, while reducing data traffic. Spectral CT image obtained by computer simulation is being presented to verify our concept.

抄録全体を表示

To analyze human gestures from movies is important for the evaluation of non-verbal communication. We make skin color automatic extraction module to analyze gestures from digital movies instead of special equipment. We evaluated the module, whose result was compared with the manual trace. The average of the over extraction rate was 19%, and the average of the under extraction rate was 21%.

抄録全体を表示

This paper presents an efficient browsing method of tennis serve videos using an acceleration sensor. First, the proposed method detects serve duration from practice video by the correlation between the right hand acceleration data in the serve motion. Next, serve duration is divided into three sections of toss, shot and follow-through using the right and left hand acceleration data. Finally, the method produces a summary image representing the motion of each three sections.

抄録全体を表示

We propose a simple correction method for perspective distortion by using the descriptive plane geometry. In this proposed method, the inverted tracing for geometry of perspective projection is used to estimate an original aspect ratio of target object. Four sides of rectangle in the object are calculated by line segment detector from the image captured on a slant, and the varnishing point is estimated from the intersected point of four sides. The constraint of right angle in target object and identified varnishing point provides us the original four corner points. By solving simultaneous equations, we can acquire the projective transform matrix for the correction of perspective distortion. Our proposed method is based on the hypothesis that target object has the rectangle shape on the plane, and center of target object fits the center of perspective transformation.

抄録全体を表示

We propose a new evalua tion method of rock condition using new fo ur directional Sobel filters to detect cracks f o r a tunnel face image captured by a widespread camera. The detected cracks show the property of the rock, and we think that the amount of the cracks are one of some objective index es of rock quality . W e confi rmed the usefulness of rock crack for sequent real tunnel face image.

抄録全体を表示

We present a noise-robust contrast enhancement algorithm using guided filter. Noise level of our method is inhibited by 36 % compared with the related work using guided filter[1][2], while visibility of our result is comparable to that of previous result.

抄録全体を表示

Because of the regularizer of optical ow, estimated motion accuracy around unreliable regions such as occlusions decreases. We propose a reliability measurement of the regions and an adaptive weight control of the regularizer based on the reliability for optical ow estimation.

抄録全体を表示

C onventional tracking method s by a drone re quire the preprocess and several restriction for tracking. I n addition, the price of dr one increases in case of using a special sensor . W e propose an automatic tracking method th at is inexpensive and less restrictive by estimating positional relati onship between a drone and a n object to track .

抄録全体を表示

We propose a disparity estimation method for stereo images with different focus positions. We exploit color line models, which represent the same characteristics independent to the focus positions, for exploring the stereo correspondence. Using synthesized stereo images, we demonstrate the effectiveness of our method.

抄録全体を表示

In this paper, we propose a novel light field imaging system for high-resolution multi-viewpoint images acquisition by using an organic photoelectric conversion film. Our experiments using synthetic images show that our method outperforms other previous methods.

抄録全体を表示

Experimental results indicate that small changes in facial expression can better be detected in “AU” (action unit) features from a Kinect sensor than by an earlier expression recognition method trained with canonical expression data.

抄録全体を表示

This paper proposes a cost-efficient palmprint identification method using convolutional sparse coding of phase information. The proposed method can reduce the computational time of POC for palmprint identification without sacrificing performance. Through experiments of palmprint identification using public databases, we demonstrate effectiveness of the proposed method compared with conventional methods.

抄録全体を表示