Acquiring absolute orientation of a movable body (e.g. robot) is desirable. Typically, this information is acquired by integrating the angular velocity from Gyroscope. Constrained by operating principles of conventional orientation sensors, they can only sense angular velocity of a body but not angle or absolute readings. Sensor error accumulation is the major drawback of incremental sensor. Drifting error is one of the most noticeable errors of gyroscope. Furthermore, the initial orientations of a body are unknown without using absolute sensor. Electric-compass is able to sense planar orientation of a body. Yet, three-axis orientation sensor is not available nowadays.

Accordingly, an absolute three-axis orientation sensing method is proposed. The concept of this method is to use visio...[ Read more ]

Acquiring absolute orientation of a movable body (e.g. robot) is desirable. Typically, this information is acquired by integrating the angular velocity from Gyroscope. Constrained by operating principles of conventional orientation sensors, they can only sense angular velocity of a body but not angle or absolute readings. Sensor error accumulation is the major drawback of incremental sensor. Drifting error is one of the most noticeable errors of gyroscope. Furthermore, the initial orientations of a body are unknown without using absolute sensor. Electric-compass is able to sense planar orientation of a body. Yet, three-axis orientation sensor is not available nowadays.

Accordingly, an absolute three-axis orientation sensing method is proposed. The concept of this method is to use vision sensor to mimic human-being to understand changes of state of orientation indicator (e.g. compass). The absolute orientation is sensed by measuring 3D relative angular positions between rotated body and fix-orientation reference body. Force balancing phenomena of liquid submerged body and magnet creates a 3D global reference body which maintains its orientation by self-regulation. This body is patterned user-defined color features on its surface and camera is employed to observe these features. Absolute 3-axis orientation information is decoded by mapping from image based on known orientation-features relations.

In this thesis, the working principle of proposed sensor is illustrated. Sensor prototype is fabricated. And, model-based orientation detection is also designed to compute three-axis orientation from image of reference body and virtual model. Similarities of captured image and images of rotating virtual model are criteria to estimate orientation. Furthermore, color classification is achieved by probabilistic estimation. Finally, the feasibility and accuracy of sensor are examined.