In a quest to make computing more interactive, Intel on Wednesday said it is working on video search technology that it hopes to bring to its future multimedia platforms.
The video search technology, which is being developed at Intel labs in U.S. and China, cuts down videos frame-by-frame and then uses image and face recognition technology to recognize faces, objects, voices, locations and movements. The frames are then patched together to make video search possible.
For example, users will be able to search videos of football games to zoom into moments when their favorite players score, said Lin Chao, a researcher with Intel. The technology recognizes and categorizes a player's face and objects like a goalpost and ball using algorithms and statistical processing technology that Intel has developed.
Once a user requests to see the goal, the technology looks for frames that contain related objects and delivers the video to the user.
Users can zoom into specific moments without watching entire videos, Chao said. The technology's recognition capabilities also help categorize images by person and object, which saves users from typing keywords to tag photographs.
However, the technology has challenges that can be overcome as processing power increases, Chao said. Processing a video to make it searchable takes hours, as current processing on PCs is limited. Chao couldn't predict when the technology would reach consumers.
The technology is part of Intel's "visual computing," which combines multiple cores, software development platforms and graphics capabilities to enable a more human interaction with a PC, said Justin Rattner, chief technology officer of Intel during a keynote at Intel's research show in Mountain View, California, on Wednesday. Intel wants to use the visual computing platform to enable interaction with a PC in life-like 3-D environments or to analyze video instantly.
Intel is already working on the Larrabee platform, which will combine multicore processors, multithreaded streams and graphics capabilities to deliver teraflops of processing power. Larrabee is due for release in 2010.
Intel's researchers are also working on a project that can track human activity to help caregivers. The company implemented a pilot in Seattle, where it has deployed monitors in 20 homes to track human activity.
RFID (radio frequency identification) tags are attached to objects such as toothbrushes, combs and medicine containers, and when those objects are moved, the tags tell an electronic monitor. RFID bracelets worn by members of the household identify who moved the object.
Ultimately, the technology could be used to identify if someone has taken medicine, for example. If the medicine container with an RFID tag wasn't opened, the monitor would alert the caregiver.
The activity tracking is unreliable for now, at between 70 percent to 90 percent accuracy, said Matthai Philipose, a researcher at Intel. It needs to reach between 95 percent to 98 percent to become reliable, he said. No study on the technology has been done to see if it is commercially viable, so it may or many not reach users in the future, he said.
Human problems could also affect the use of this technology, Philipose said. In countries like India or China, where homes have multiple caregivers, the workload of monitoring tasks will need to be broken up equally. If multiple people monitor tasks ineffectively, for example, people may end up taking medicine twice.
At this stage in the project, the RFID tags are protected by black blobs of plastic and so the lack of pleasing aesthetics is a problem, Philipose admitted. However, the plastic makes the RFID tags dishwasher safe, he said.
The video search technology, which is being developed at Intel labs in U.S. and China, cuts down videos frame-by-frame and then uses image and face recognition technology to recognize faces, objects, voices, locations and movements. The frames are then patched together to make video search possible.
For example, users will be able to search videos of football games to zoom into moments when their favorite players score, said Lin Chao, a researcher with Intel. The technology recognizes and categorizes a player's face and objects like a goalpost and ball using algorithms and statistical processing technology that Intel has developed.
Once a user requests to see the goal, the technology looks for frames that contain related objects and delivers the video to the user.
Users can zoom into specific moments without watching entire videos, Chao said. The technology's recognition capabilities also help categorize images by person and object, which saves users from typing keywords to tag photographs.
However, the technology has challenges that can be overcome as processing power increases, Chao said. Processing a video to make it searchable takes hours, as current processing on PCs is limited. Chao couldn't predict when the technology would reach consumers.
The technology is part of Intel's "visual computing," which combines multiple cores, software development platforms and graphics capabilities to enable a more human interaction with a PC, said Justin Rattner, chief technology officer of Intel during a keynote at Intel's research show in Mountain View, California, on Wednesday. Intel wants to use the visual computing platform to enable interaction with a PC in life-like 3-D environments or to analyze video instantly.
Intel is already working on the Larrabee platform, which will combine multicore processors, multithreaded streams and graphics capabilities to deliver teraflops of processing power. Larrabee is due for release in 2010.
Intel's researchers are also working on a project that can track human activity to help caregivers. The company implemented a pilot in Seattle, where it has deployed monitors in 20 homes to track human activity.
RFID (radio frequency identification) tags are attached to objects such as toothbrushes, combs and medicine containers, and when those objects are moved, the tags tell an electronic monitor. RFID bracelets worn by members of the household identify who moved the object.
Ultimately, the technology could be used to identify if someone has taken medicine, for example. If the medicine container with an RFID tag wasn't opened, the monitor would alert the caregiver.
The activity tracking is unreliable for now, at between 70 percent to 90 percent accuracy, said Matthai Philipose, a researcher at Intel. It needs to reach between 95 percent to 98 percent to become reliable, he said. No study on the technology has been done to see if it is commercially viable, so it may or many not reach users in the future, he said.
Human problems could also affect the use of this technology, Philipose said. In countries like India or China, where homes have multiple caregivers, the workload of monitoring tasks will need to be broken up equally. If multiple people monitor tasks ineffectively, for example, people may end up taking medicine twice.
At this stage in the project, the RFID tags are protected by black blobs of plastic and so the lack of pleasing aesthetics is a problem, Philipose admitted. However, the plastic makes the RFID tags dishwasher safe, he said.