| NSF Org: |
IIS Division of Information & Intelligent Systems |
| Recipient: |
|
| Initial Amendment Date: | September 12, 2013 |
| Latest Amendment Date: | September 12, 2013 |
| Award Number: | 1317947 |
| Award Instrument: | Standard Grant |
| Program Manager: |
Jie Yang
jyang@nsf.gov (703)292-4768 IIS Division of Information & Intelligent Systems CSE Directorate for Computer and Information Science and Engineering |
| Start Date: | September 15, 2013 |
| End Date: | August 31, 2015 (Estimated) |
| Total Intended Award Amount: | $433,711.00 |
| Total Awarded Amount to Date: | $433,711.00 |
| Funds Obligated to Date: |
|
| History of Investigator: |
|
| Recipient Sponsored Research Office: |
3451 WALNUT ST STE 440A PHILADELPHIA PA US 19104-6205 (215)898-7293 |
| Sponsor Congressional District: |
|
| Primary Place of Performance: |
PA US 19104-6205 |
| Primary Place of
Performance Congressional District: |
|
| Unique Entity Identifier (UEI): |
|
| Parent UEI: |
|
| NSF Program(s): | Robust Intelligence |
| Primary Program Source: |
|
| Program Reference Code(s): |
|
| Program Element Code(s): |
|
| Award Agency Code: | 4900 |
| Fund Agency Code: | 4900 |
| Assistance Listing Number(s): | 47.070 |
ABSTRACT
With advances in camera technologies, and as cloud storage, network bandwidth and protocols become available, visual media are becoming ubiquitous. Video recording became de facto universal means of instruction for a wide range of applications such as physical exercise, technology, assembly, or cooking. This project addresses the scientific and technological challenges of video shooting in terms of coverage and optimal views planning while leaving high level aspects including creativity to the video editing and post-production stages.
Camera placement and novel view selection challenges are modeled as optimization problems that minimize the uncertainty in the location of actors and objects, maximize coverage and effective appearance resolution, and optimize object detection for the sake of semantic annotation of the scene. New probabilistic models capture long range correlations when the trajectories of actors are only partially observable. Quality of potential novel views is modeled in terms of resolution that is optimized by maximizing the coverage of a 3D orientation histogram while an active view selection process for object detection minimizes a dynamic programming objective function capturing the loss due to classification error as well as the resources spent for each view.
The project advances active sensing and perception and provides the technology for further automation on video capturing. Such technology has broader impact on the production of education videos for online courses as well as in telepresence applications. Research results are integrated into robotics and digital media programs addressing K-12 students.
PUBLICATIONS PRODUCED AS A RESULT OF THIS RESEARCH
Note:
When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external
site maintained by the publisher. Some full text articles may not yet be available without a
charge during the embargo (administrative interval).
Some links on this page may take you to non-federal websites. Their policies may differ from
this site.
PROJECT OUTCOMES REPORT
Disclaimer
This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.
When searching for an object in an image, traditional as well as modern approaches slide a filter window per object part over the entire image. Biological systems rather use an attentional scan path based on extracted saliency in the image. The main result of this project has been a novel approach for object detection that actvely selects where to spend computational resources.
Object detection is regarded as a planning problem where the policy determines which object part filter should ne applied next and where. Such a policy should minimize the number of filter operations and simultaneously the expectation of the misclassification error. Finding such a policy can be reduced to a dynamic program.
Applying this active selection strategy of where to apply part filters has resulted in acceleration between 5 and 30 times compared to the sliding window approach without sacrificing precision and recall.
In a second application of active perception, a move selection policy has been studied in the context of absolute semantic localization. Given landmarks on a known map, a new control scheme has been devised that makes a robot move in such a way that the error in localization and the energy spent in the moves is minimized.
Last Modified: 02/01/2016
Modified by: Kostas Daniilidis
Please report errors in award information by writing to: awardsearch@nsf.gov.
