To achieve an active optical filter for television cameras, we created a prototype of a metal salt precipitation-type electrochromic dimming element. We found that the transmittance could be reversibly varied in response to an applied voltage, and that flat optical transparency could be obtained across the entire visible light region.

抄録全体を表示

This platform is made up of storage virtualization technology, contributing to the improvement of News & Sports workflow efficiency. This time, I will detail the devising and practical application of this platform.

抄録全体を表示

Since Hybridcast Service s has started in September 2013 , b roadcasters and a pp lication d evelopers have straggled to develop H ybridcast applications and overcome the differences of the execution performance of each browser for various H ybridcast applications on each TV receiver . In order to help that development, we have developed some prototype Hybridcast benchmark test applications to evaluate Hybridcast HTML5 browser performance on each receiver . This pap er describes the approach to design the Hybridcast benc hmark test applications .

抄録全体を表示

When we get Convol utional Neural Network by transfer learning , this network has some filter that raise incorrect recognition rate . T his paper propose s a method to select optimal filter and remove unnecessary filter .

抄録全体を表示

This paper propose s a robust automatic extraction method of mouth area from facial image by implementing the Convolutional Neural Network (CNN) - based classification of each pixel. This will help to draw expressive portrait automati cally over wide - ranging facial images. I t is possible to expand the variety of communication.

抄録全体を表示

Road signs provide important information to guide and regulate the behavior of drivers and pedestrians to make their journeys safer and more comfortable. In this study, we propose a method to recognize the speed-limit signs using pattern matching for night scene videos.

抄録全体を表示

We present a method to extract feature vector s from Manga image s by CNN for sketch - based Manga retrieval . We show ed that fine - tuning Ale xNet pre - t rained on ImageNet using a large amount of Manga face image s from Manga109 improves accuracy of Manga character retrieval .

抄録全体を表示

This research proposes a method for generating distributed representations, or vectors, of food names in food recording systems, which loyally represent the characteristics of the dishes that the names are referring to. These vectors can be used to realize a flexible and expandable associative search system in such services.

抄録全体を表示

We present a new method of novelty detection in image recognition based on convolutional neural network (CNN). We use Sigmoid Layer as the last layer of a CNN instead of Softmax Layer. As a result, we discovered that a CNN with Sigmoid Layer can detect novelties in an easy dataset better than that with Softmax Layer, but worse in a difficult dataset.

抄録全体を表示

W e propose a method of point cloud thinning for extracting core of reinforcing steel r ods from point cloud data. The method thins point cloud data using a density function field and extract the core lines . An experiment result shows that the method is effective.

抄録全体を表示

It is difficult for a perso n to coord inate the connection between CG motions and takes time even if we use mo tion capture systems. this paper proposes a method of automa tic generation of the c onnection between CG motions for improvement of the work efficiency and generation of more realistic CG motions .

抄録全体を表示

This paper aims to create frame-sampled and stabilized omnidirectional video. We propose a method which considers camera motion between omnidirectional frames. The results show that our method can select suitable frames for stabilization in omnidirectional video.

抄録全体を表示

Projection mapping is a useful technique to control material perception. In this study, we propose a method to control texture appearance by a frequency-based projection mapping. We conducted subjective evaluation experiments to validate the feasibility of our method. The experimental results show out method is effective for texture appearance control.

抄録全体を表示

This paper proposes three-dimensional measurement system for foot using RGB-D cameras. We get threedimensional point cloud data using four RGB-D cameras and align these data to get three-dimensional measurement data. We aim at estimate best fit class between shoes and feet data using each three-dimensional features.

抄録全体を表示

We are considering a new style of watching television with a communication robot in the living room. In this paper, we propose a detection method of TV position combining frame subtraction and edge detection from camera images of the robot. Experimental results in an environment simulating a living ro om show the effectiveness of our proposed method.

抄録全体を表示

We conducted subjective evaluation experiments to clarify an appropriate presentation in terms of display size. The results show that viewers have str onger impressions of the object s when they are shown on the display as their actual size.

抄録全体を表示

W e conducted subjective tests to clarify a relation ship between the characteristics of the display brightness and impression s . We found that the display brightness had an effect to enhance some impressions.

抄録全体を表示

We propose a n image processing mode l to enhance impressions of image s . We conducted a preliminary experiment to design model parameters. We then confirmed the model works through the subj ective evaluation experiment on the processed images.

抄録全体を表示

We conducted subjective tests to clarify a relation between Wide-Color-Gamut(WCG) and impression. We found that WCG images had an effect to enhance some impressions.

抄録全体を表示

We have been evaluating the driving behavior of elderly drivers by the TTC(Time To collision). However, evaluation was that differ with the insp ection decision. In the study, w e developed a new evaluation method for obtaining the same result as the in spection decision automatically.

抄録全体を表示

While we operate a veh icle, divided attention is one of an important factor. Using a driving simulator, w e study about an estimation of d river ’ s divided attention.

抄録全体を表示

Conventional writing-type executive functional tests are so complex that elderly people can not easily undergo the tests. Thus we made an automated evaluation system of executive function. In this report, introduces the system and verifies its usefulness.

抄録全体を表示

We have conducted an experiment al shoot ing of a soccer ball with HDR (High Dynamic Range ) imaging method to improve accuracy of extracting moving objects for locus analysis . The results showed effective ness of HD R imaging method.

抄録全体を表示

This paper describes development and practical use of a very small sized camera which can be attached to sports referee and capture “the referee's eye” for live broadcasting. The camera has HD-SDI(1080/59.94i) output at light weight (40g),very small size (45x26x26mm) and has been in practical use for live sports broadcast from 2015.By this development, the viewers came to be able to enjoy a reality picture of the referee's glance.

抄録全体を表示

An 8K slow-motion system is strongly desired for 8K sports content production. We have developed a 240-fps test shooting system using full-featured 8K equipment and implemented a slow motion shooting experiment of sport contents.

抄録全体を表示

We developed uncompressed 4K player and recorder with th e goal of 4K broadcasting on BS channels in 2018 .The device we developed is composed of four uncompressed HD players and recorder s with SSD s . The cost of storage media is less than 10 percent of e xisting devices because of using universal SSDs.

抄録全体を表示

This paper proposes a n AR sightseeing s ystem suitable for a sightseeing spot with proximity detection and markerless AR . The system uses the beacon for the proximity detection of the user and AKAZE features for the markerless AR . Using these techniques realize the h ighly precise system which do es not spoil a scene .

抄録全体を表示

In the displaying virtual objects through the glass such as a show window , i t seems to exist in front of the glass because of the reflect ion of the ambience. T his paper proposes extracting reflections using stereo camera and displayi ng as behind on the basis of the reflections .

抄録全体を表示

An invisible AR system using polarized light doe sn't spoil object’s appearance. However, there are still issues in invisibility an d recognition accuracy of AR ma rker. This paper propo ses a solution of these issues and reports the improvement of an invisible AR system using pol arized light.

抄録全体を表示

This paper proposes an interactive AR exhibitions using a pr ojection mapping for exhibition objects which have textures and colors . T extures and colors of the exhibition objects oft en disturb projection mapping. We try to s olve this problem with projector - camera system.

抄録全体を表示

Our novel studio robot helps to improve TV program production featuring live performers and CG counterparts. It has a light sensor and a motion sensor enabling more integrated video compositing with lively interaction between the performers and CG characters.

抄録全体を表示

We accelerate a view-dependent texture mapping process to generate novel view images for stereoscopic HMD from captured images. By using the similarity between a stereoscopic pair of HMD images we are able to improve the performance of the VDTM process.

抄録全体を表示

It is important to know about surrounding objec ts and beware steps when wheelchair users move around. However it is diff icult to get these information because of the lower eye's height and restrict physic al movements than normal people . This paper proposes a method of providing surround information usi ng 3D information getting from wheelchair users above.

抄録全体を表示

If an input method for physical disabilities uses eye-blink as input decision, it requires classification of eye-blink types. In this research, we were corroborated significant difference in duration time for each type of eye-blink. Based on this, we developed auto identification method for voluntary eye-blink. In this method, it calibrate on every examinee. We report auto identification method and the experiment of input system using conscious eye-blink.

抄録全体を表示

This report propose a message board “Resonant Bits“, which shows input message draw n on it by the user selectively b ase d on frequencies of the voice as if the message resonantly emerges responding to the us er’s voice.

抄録全体を表示

This paper proposes an interactive system “Sound of Surrounds” , which you can virtually place any sound in a real scene around you and instantly replay it by directing th e position associated by the cue in the scene.

抄録全体を表示

The research analyzed the dialogue behaviors of movie appreciator pairs. The dialogue scene of the subject pair was shot from 1st, 2nd and 3rd - person views by video cameras. The recorded voice was converted to text data. The analyzed data of different pairs was compared to extract behavior features.

抄録全体を表示

A stereoscopic vision method using a single still picture is proposed; the method is based on object background and foreground estimation. When two object s are overlap each other , the foreground object hides the background object . We compute the c onvex hull of the each object, and w e estimate which object is foreground or background in the image. After the foreground or background estimation, stereoscopic vision is performed using the anaglyph method . The experimental results show that the method is realized using a real image.

抄録全体を表示

A n e ndoscopic surgery training system using a full HD glass - less stereosco pic display based on time - division multiplexing parallax barrier is realized. By using a stereoscopic camera, stereosc opic live action movie is displayed with our system. The result of the experiment shows that the operators can work faster under the 3D co ndition.

抄録全体を表示

A full HD aerial stereoscopic display is realized by combining a convex lens and time-division parallax barrier. Undistorted stereoscopic images are presented by the parallax barrier system with correction of image distortion.

抄録全体を表示

A thin and wide HUD is attainable by use of a n autostereoscopic display . Crosstalk of left - eye image and right - eye image, however, can cause serious trouble in practice. In this paper w e evaluate the acceptable le vel of crosstalk for 3D HUDs.

抄録全体を表示

We have developed the super multi - view head - up display (SMV - HUD) for automobiles . In this paper, we evaluate d the temporal characteristics of the depth perception for the three - dimensional images produced by the SMV - HUD . The accuracy of the depth perception was measured for different observation time , such as, 0.5, 1.0, and 2.0 s .

抄録全体を表示