Authors: [tex2html_wrap4228]M. El-Gamal, M.D. Levine, G. Roberts
Investigator username: levine
Subcategory: sensor and processor design
Significant computational power is necessary to guide and control a mobile robot based on sensory data. Powerful, compact, and modular signal processing systems that can be mounted on the robot are needed in order to ensure its autonomy. Very Large Scale Integrated (VLSI) circuits have reached the point where they can offer potential solutions to these requirements. We are interested in the design and implementation of VLSI systems that will provide a mobile robot with visual information in real-time in order to accomplish specified tasks.
Combining photo-transduction and signal processing at the sensor level is an emerging approach in the development of VLSI systems for computer vision. This dramatically reduces the data bandwidth that must be processed by subsequent stages. Such an approach is supported by biological systems in which data processing at the retinal level results in a very large reduction in the information transmitted for further visual processing.
CMOS technology will be used to implement the photosensitive elements, and naturally an analog preprocessing stage of much lower output bandwidth than the input image will follow. Despite their high speed , low power, and small area, analog systems suffer from lack of flexibility and storage capacity. This tends to limit their use to early vision tasks such as image filtering, edge detection and moment extraction.
The output of this stage will evidently need further processing to achieve higher level tasks (e.g. target tracking). The stability, accuracy, and programmability of digital systems make them better candidates to achieve these processes. Therefore, the analog stage will be followed by an analog-to-digital conversion circuit. This will result in a large digital data flow that can lead to a bottleneck at the chip boundary. We believe that a digital processing stage on chip, where the communication bandwidth is still potentially high, can further reduce the amount of data to be transmitted from the chip. This data will then be used to compute a measure of the symmetry in the scene.