An Intelligent Autonomous Vision Device

The article describes the concept of an intelligent autonomous vision architecture. The main idea is to have a compact, i.e., slightly larger than a cigarette-pack, device which combines the basic features of a camera, a computer, and a controller. This means the device has downloadable functionality to read an image from a vision-chip, to process it onboard, to trigger and/or control external devices, and to transmit data if necessary. It is intended as inexpensive and extremely compact alternative to the camera-framegrabber-PC approach in applications with low to medium processing needs. The device can for example be used for surveillance. It can be programmed to trigger an alarm and a recording-function based on a change in diierence images, which indicates an intrusion in the observed area. Similar and even more complex functions can be used for applications within e.g. robotics, production, and new forms of HMI. Concretely, the design-scheme of VisionCube is presented. It is based on a FUGA15 C-MOS image chip and a Motorola MC68332 micro-controller. VisionCube can be interfaced to other devices via sixteen binary I/O channels, three motor-controllers, two RS232 ports, and a radiolink. The motor-channels can be used to control a pan-tilt unit, allowing active vision. Additional eight analog inputs can be used for sensor-fusion. The total cost of parts is less than 800$.