Design of an electronic saccadic imaging system

Abstract

An electronic implementation of a saccadic vision system consisting of a CMOS image sensor ("the eye'y mounted on a mechanical servomotor ("fhe neck'y is presented. The camera first surveys the wide-angle scene at low-resolution ($A XGA), and then switches to a smaller, higher-resolution (SXGA) subwindow for more detailed inspection of salient areas. Image processing on the low-resolution scene quickly extracts the locations of objects of interest against a cluttered background. A saliency map is then generated to plot the regions of interest that require the attention of the high-resolution fovedsubwindow.Saccadic jumps are performed electronically by the camera to quickly redirect the subwindow between salient points.In this paper we discuss the design issues involved in the construction of our system protorype (including image processing strategies employed fa achieve realtime feature extraction), and we present the preliminaiy results of our tests using theprototyge.

Calibration of a saccadic camera system to adapt to lens distortions

“…In our previous work, an efficient tracking system was demonstrated using foveated vision to increase the performance of object tracking and resource allocation [17]. In this work, a method to apply the saccadic camera to lens distortion compensation is discussed, using a known mechanical state to estimate methods of handling issues arising from the image formation.…”

“…In the previous work [17], we discussed the advantages of the electronic reconfigurability of the sensor: fast and accurate repositioning of the foveal subwindow for efficient target tracking and investigation. Let us now discuss some benefits of the mechanical component of the system, and apply it to some fault tolerance applications.…”

“…The first four features in this list are detected during scan mode. Motion is detected separately in saccade mode and is discussed in our previous work [17]. During scan mode, the survey image is processed using a 'preattentive' algorithm to extract the salient features and to generate the saliency map, which indicates the coordinates of potential regions of interest to be revisited in saccade mode.…”