Object recognition from multiple viewpoint images

Estimated a suitable camera arrangement for object recognition and pose estimation from multiple viewpoint images.

Object recognition from multiple viewpoint images Object recognition from multiple viewpoint images

Recognition of low-resolution characters (symbols) captured by camera-equipped phone

Developed a method to recognise low-quality characters using multiple frames of video data. Developed a training method that takes the actual degradations into consideration.

Recognition of low-resolution characters captured by camera-equipped phone Recognition of low-resolution characters captured by camera-equipped phone Recognition of low-resolution characters captured by camera-equipped phone

Person identification from low-resolution surveillance video

Developed a method to recognise persons (faces) from low-quality surveillance video using super resolution technique.

Recognition of low-resolution characters captured by camera-equipped phone

Construction of a Street Image Map using car-mounted omni-directional camera

Constructed a Street Image Map database, which integrates images and information from GPS for car navigation systems.

Construction of a Street Image Map using car-mounted omni-directional camera Construction of a Street Image Map using car-mounted omni-directional camera

Estimation of human performance detecting objects

When driving a vehicle, a driver recognizes various objects, such as traffic signs, traffic signals and pedestrians. We are developing a method for estimating the human performance detecting objects (detectability or visibility) based on image processing and pattern recognition techniques. By using the visibility estimation technique, we are trying to develop next advanced driver assistance systems (ADAS) considering driver’s condition.

Detectability of pedestrians Visibility of traffic signals

Recognition of various objects in images captured by car-mounted camera

We are developing technologies to support drivers using car-mounted camera by recognizing the weather condition, traffic signs, pedestrians, etc. To improve the performance, we proposed an environment adaptive detector. In addition, various types of generative learning methods have been proposed to reduce the cost for constructing the detector.

Recognition of various objects in images captured by car-mounted camera Recognition of various objects in images captured by car-mounted camera

Fast detection and application of near-duplicate video segments in a long video

We are working on the detection of all near-duplicate video segments in a long video, together with its application to related news detection regardless of languages and analysis of TV program structures and detection of commercials.

Fast detection and application of near-duplicate video segments in a long video Fast detection and application of near-duplicate video segments in a long video

News video understanding

We are working on efficient browsing of a large amount of news video by focusing on social networks of people that appear in news shows, or tracking the development of news topics. Such contents are analyzed through a multimodal approach using both text and image information.

News video understanding News video understanding

Contents analysis for supporting cooking activities

We are working on the analysis of motions in cooking videos and recipe texts aiming to provide visual information that should support cooking activities.

Contents analysis for supporting cooking activities