Various machine-vision applications like shape detection and object classification, object orientation monitoring, measurement tasks, etc. do not rely on conventional image data but benefit from precise information about the edges of usually fast moving objects.

Using conventional clocked line-scan sensors in high-speed machine vision applications results in large amounts of highly redundant image data of which a substantial fraction do not provide any information necessary to accomplish the task or even to increase reliability or precision.

The dual-line sensor chip uses a pixel circuit that operates autonomously and responds with low latency to relative illumination changes by generating asynchronous events. The events are used to trigger the transmission of a data packet containing the active pixel address. Time-stamps are combined with the corresponding address events to compose a synchronous stream of data packets and are read out via a 3-stage pipelined bus arbiter.

• The output data volume of the dual-line sensor chip depends on the dynamic contents of the target scene and is typically orders of magnitude lower than the data output produced by conventional clocked line sensors for equal temporal resolution and information content.
• The sensor performs precision time-stamp assignment at the pixel level with sub-microsecond temporal resolution.
• The combination of event-driven pixels and the time/velocity measurement capability of the dual-line configuration completely eliminate the need of synchronizing a readout frequency to the target object speed.
• The dual-line arrangement offers additional functionality such as measurements of trajectory angles by correlating the timed address-event streams from the two pixel lines.
• The pixel circuit used in this sensor is able to detect contrast changes of a few percent over a dynamic range of > 120 dB, from under 100 mlux to more than 100 klux of scene illumination (f/1.4 lens) at low latency. This feature qualifies the sensor for use under difficult, variable lighting conditions e.g. in outdoors applications like road traffic monitoring.
• The sparse data delivered by the sensor require low computational effort and power consumption in the post-processing stages and allow for compact and low-cost embedded or mobile systems.

The figures show sensor data in response to the black bar stimulus depicted in (a). The stimulus pattern consists of bars with orientations varying by 0.25 degrees. The pattern was moved with a speed of 5.8m/s at a distance of 15 cm from the lens (f/1.6, 8 mm), yielding a field-of-view of ~12 cm diameter.

Figure (b) shows event output from one sensor line plotted as pixel address vs. time and (c) angle histograms of the leading edges of each stimulus bar. The angle histograms are generated in the style of an abscissa-projected Hough transform. The results indicate that the sensor is able to resolve angle differences of 0.25 degrees or better under these conditions, proving the sensor`s temporal resolution.

The pixel dynamic range of > 120 dB was verified using neutral density filters (0 to ND5) placed between the sensor and the stimulus illuminated at 25 klux. The event output and angle histograms from one stimulus edge for 6 decades of illumination are shown in (d) and (e) respectively. The data indicate that the angle resolution of (c) can be maintained over 6 decades of illumination.