ITEE Innovation Expo 2008 : Project Details

FLY-EYE SENSOR FOR VISUAL ODOMETRY

The estimation of self motion (egomotion) using visual information alone has been a major area of research over the past decades. A new trend in the development of navigation and obstacle avoidance systems for miniature aerial vehicles has been to analyse how insects perform this function. This thesis describes the design, construction and evaluation of a wide-angle optical flow sensor based upon biologically-inspired image motion detection.

The calculation of egomotion using image motion can be vastly improved by a wide-angle or omnidirectional image sensor. A larger field of view allows for more accuracy in determining total motion, since translational and rotational components are more easily separable.

A planar arc of sensors was used a one dimensional prototype in order to perform some initial investigation of suitable devices and parameters for the hemispherical array. It was found that the integrated angle of the calculated optical flow yielded a strong correspondence with the absolute angle measured from an encoder, validating the hardware selection and optical flow algorithms.

For the hemispherical sensor a section of a truncated icosahedron was chosen to provide a structure upon which 116 phototransistors could be uniformly positioned. As the sensor is configured as a number of pentagons and hexagons connected to form a polyhedron, it was decided to base the circuit board construction on these shapes. Each sensor is connected via a series of analog multiplexers to the ADC port on an Atmel AVR microcontroller and communication to the PC is performed using a USB connection.

The egomotion results will be compiled in the coming weeks using a robot arm to create a known motion that can be used as ground truth for evaluation of the sensor performance.