Feature Extraction Through Time

Robots that perform repetitive tasks without human intervention have been used for decades in automated manufacturing. These systems are programmed to perform simple repetitive tasks with little or no ability to sense or react to changes in their environment. Collisions and other undesirable events are prevented by careful control of their work area. If a collision does occur it is problematic, but work can usually continue after an investigation, the loss of a few parts and some repairs. Robotic collisions in space do not follow this model. The consequences of a collision for a robotic system during a space flight can be catastrophic. Even a minor collision between the robot and an object in space could cause a system failure, the loss of the mission or even the death of an Astronaut. There have been many suggested means for avoiding collision in space. One approach for solving this problem is to use real-time vision data for authenticating synthetic models used in collision detection. There are many techniques for analyzing digital images, however, space based imagery is constrained by a number of technical challenges including challenges related to imaging, lighting variation, object appearance and system constraints.

57th International Astronautical Congress, IAC-06-B4.4.03, Valencia, Spain Oct 2-6, 2006 (in press)

Feature Extraction Through Time