PROBABILISTIC REAL-TIME DOMAIN AWARENESS LEVERAGING COMPUTER VISION AND COMPUTATIONAL INTELLIGENCE
Open Access
- Author:
- Bolden, Mark Patrick
- Graduate Program:
- Aerospace Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- February 07, 2017
- Committee Members:
- David Spencer, Thesis Advisor/Co-Advisor
- Keywords:
- domain awareness
space domain
space debris
real-time
debris tracking
computer vision
computational intelligence
tactical object tracking
clustering
break-up event tracking
debris cloud tracking - Abstract:
- In tactical operations, decision making must exceed the speed of events. In the space domain, events, such as a satellite break-up event, can occur unexpectedly creating large fields of debris objects. These debris objects are uncontrolled and pose an immediate threat to other satellites. Satellite operators must quickly decide whether or not to maneuver their satellites. If a maneuver is necessary, the operator must also decide where to maneuver the satellite to avoid a probable collision. Due to the complexity of this challenge, traditional statistical approaches struggle to fuse information to track debris clouds on a timeline that exceeds the speed of events. This thesis presents an alternative approach to statistical filters for real-time debris cloud tracking. The approach produces the same level of accuracy in the same amount of time across the entire domain independently from the number of objects in the domain. The approach relies on a new computer vision transform coupled with a new computational intelligence cluster detection algorithm. Combining these techniques enables real-time (millisecond) updates using information from any modality where the uncertainty can be approximated or bounded, ranging from sensing to anecdotal sources. It is capable of ingesting observations that do not meet the typical observability criterion, to include negative detections. It provides real-time state estimates in addition to a full domain population distribution assessment. This includes knowledge of where there is an object, where there is not, and where there is not enough evidence to make a determination. Ultimately, the approach enables the scalable real-time domain awareness required for a decision maker to understand the domain risks and the uncertainties for tactical decision-making. More background on space debris is discussed as the motivation for the technical challenges tackled by this research. Applying computational intelligence with computer vision is presented as a novel approach to solving these challenges in real-time. The concept is first explained on a simple example that leverages a standard computer vision technique known as the Hough transform. Next, the hardware and software design leveraged for testing is described along with performance results on the simple one-dimensional tracking example. How to apply this technique to the space domain is discussed in detail with a proof of concept for the computer vision transform. Potential applications of the computational intelligence algorithm to other domains are discussed. The technique appears to be a viable alternative to statistical filter approaches with significant theoretical advantages, however no direct comparison is presented in this thesis.