We implemented a novel event-based vision sensor with two operation modes: 1) intensity mode and spatial contrast detection. They can be combined with two different readout
approaches: 1) pulse density modulation and time-to-first spike. The sensor is conceived to be a node of an smart camera network made up of several independent an autonomous nodes that send information to a central one. The user can toggle the operation and the readout modes with two control bits. The sensor has low latency (below 1 ms under average illumination conditions), low power consumption (19 mA), and reduced data flow, when detecting spatial contrast. A new approach to compute the spatial contrast based on inter-pixel event communication less prone to mismatch effects than diffusive networks is proposed. The sensor was fabricated in the standard AMS4M2P 0.35-μm process.The sensor is conceived for networks with distributed cameras. The user can trade between image quality and the amount of data required to encode images. Some opeeration modes have very reduce output data flow. The sensor can reconfigure itself on demand to provide images with more quality if the operator needs them. It is possible to operate continuosly event when the sensor switches operation and/or reaout modes.
To compute the spatial contrast, there is a competition between the activity of the pixel and its neighborgs. Pixels spike with a frequency that is proportional to light intensity. Thus there is a lignt-to-frequency conversion inside each pixel. Pixel spikes charge a capacitor every time that they are elicit. On the contrary, neighborgs discharge this capacitor (C2 in the figure). If the capacitor voltage reaches a certain value, an event will be fired, indicating that the pixel illumination is higher than the average illumination in its surroundings. Hence, there is spatial contrast. Each pixel is connected to four neighborgs. If we disable the influence of the neighborhood, each pixel will spike with a frequency propotional to illumination. Therefore, there are two possible operation modes: light-to-freqcuency conversion and spatial contrast detection.
On the top figure, we can observe the transient voltage at the capacitor C2. The pixel provoke voltage increments. Its neighborgs decrement the voltage at the capacitor whenever fire. The voltage can only reach a programmable voltage threshold if the pixel illumination is higher than the average illumination in its neihborhood.
In adition to the two operation modes, there are two readout modes. In the nominal one, all the events are arbitered and sent out the chip. In the second one, pixels can only spike once. This is the Time-to-first-spike (TFS) readout. To implement it, only an extra transistor Mrf, a capacitor Crf, and a reset transistor are required.
The main advantage of the sensor is that the user can toggle freely between the operation and/or the readout modes. The sensor has not to be reset to do so. Thus, it can operate continuosly. Only two control bits are required to switch between the operation and the readout modes. On the bottom, we show the circuitry to control the operation and the readout modes:
Several snapshots of natural scenes are shown below. We used the nominal event readout and the two different operation modes: intensity and contrast detection.
The next figure displays the sensor operating with the TFS readout mode, and the intensity operation mode. This setting econdes images with a very reduce output data flow. The oeprator can decide how many events can be used to reconstruct the visual scene.
In the video displayed below, we can see the sensor operation modes . We also show how operation/readout modes can be toggled without reseting the sensor.
The following table summarizes the main sensor features:
Selected Project Publications
Back
