Improved dynamic shape representation using a biologically-inspired vision sensor with a synchronous arbitration

Abstract

Neuromorphic temporal contrast vision sensors are sensitive to relative intensity changes. These sensors can be exploited to detect scene dynamics and representing the resulting dynamicpsilas shapes. Moreover, these sensors are ideal for ultra-high-speed vision with low computational effort. Two aspects have been ignored within the initial conceptual design of this kind of sensors: the preservation of the high temporal resolution of the pixelspsila data and handling high peak rates. In other words, timestamping the pixelspsila data and the minimization of the data loss in case of pixel spiking at higher rate have not been intensively investigated. This work provides an on-chip solution using a synchronous Address-Event interface for maintaining the precise temporal information and reducing the data loss for high-speed applications.