Scheduling image recognition tasks based on task dependency and phase
US8973010B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | May 28, 2010 |
| Grant date | Mar 3, 2015 |
| Priority date | — |
| Expiry date | May 22, 2032 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06V2201/03
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
Embodiments of the present invention are directed to techniques for providing an environment for the efficient execution of recognition tasks. A novel environment is provided which automatically and efficiently executes a recognition program on as many computer processors as available. This program, deconstructed into separate tasks, may be executed by constructing a dependency network from known inputs and outputs of the tasks, applying project planning methods for scheduling these tasks into multiple processing threads, and dynamically assigning tasks within these threads to processors. Therefore, an efficient schedule of tasks to complete a recognition program can be created and executed automatically, for any type of recognition problem. The system will not only allow for the ability to leverage multiple processors for efficiently generating variable and customizable automatically created schedules, but will also still maintain the flexibility to use serial programming in recognition algorithms for individual objects, properties, or features.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.