An analog VLSI model of the fly elementary motion detector
NIPS '97 Proceedings of the 1997 conference on Advances in neural information processing systems 10
A Robust Analog VLSI Reichardt Motion Sensor
Analog Integrated Circuits and Signal Processing
Neuromorphic models of visual and motion processing in the fly visual system
Neuromorphic models of visual and motion processing in the fly visual system
Image and Vision Computing
| Hi-index | 0.00 |
Previous hardware implementations of the Reichardt motion detection model based on the fly visual system suffer from drawbacks such as large pixel area and fixed spatial frequency response. In this work, we introduce a two-stage implementation of the Reichardt model in order to overcome these issues. We perform the pixel-level temporal filtering required for motion processing along with photoreception on a single analog chip using continuous-time low-order filters. The rest of the computation—spatial filtering and correlation—is performed on a subsequent discrete-time digital processing stage. We designed and tested an integrated CMOS imager with integrated, pixel-level temporal filtering necessary for motion detection. This temporal filtering enhances transients and provides a delayed and non-delayed channel for each pixel in the array. By varying bias currents on the chip, we are able to control the time constant of continuous-time low pass filters that set the delay and hence the temporal response of the imager. By varying a reset switching frequency, we can control the high pass filter time constant which determines the time course of transient enhancement. Operation primarily in weak inversion results in a low-power sensor well suited for integration into a motion processing system. We have constructed a 16 × 16 temporal processing array in a 0.5 μm CMOS process. Motion information is extracted by post processing the imager output in a discrete time domain, which validates the functionality of the temporal filters.