GENX320 Camera Module

About

The GENX320 Camera Module comes with the Prophesee GENX320 Event Camera. However, you must connect it with the H7 Plus or RT1062 to use the camera module.

Please see examples for the GenX320 sensor in OpenMV IDE under File->Examples->Cameras->Event Cameras->GenX320.

For any other scripts, you need to make sure the pixformat in sensor.set_pixformat() is sensor.GRAYSCALE and that the frame size in sensor.set_framesize() is sensor.B320X320.

The Prophesee GenX320 camera module brings event-based vision sensing to the OpenMV Cam platform.

Inspired by the human retina, each sensor pixel responds independently and asynchronously to relative changes in illumination leading to efficient, high-speed and low power vision sensing.

• Efficient – only changes in the scene are sensed leading to low data rates even for fast motion. No more static background removal required.
• High-speed – no exposure time, no motion blur, delivering event histograms at 375Hz+ with OpenMV Cam H7 plus.
• Low-power – sensing only scene changes gives increased battery life with sensor consumption starting at 3mW streaming data.
• Extended Dynamic Range - Achieve high robustness even in challenging lighting conditions with up to 140dB dynamic range. Sensing only relative illumination changes allows to see changes in both dark and bright zones of a scene. No need for auto-exposure.
• Low light sensing - Sometimes the darkest areas hold the clearest insights. Use the GenX320 module to sense into twilight light levels (<5 lux) even with brightly lit zones in the FoV thanks to the sensor low-light cut-off of 0.05 lux.

Comprehensive sensor configuration options are made available right from the OpenMV IDE. Harness the full power of event-based sensing using GenX320’s built-in settings. For further information, please visit docs.prophesee.ai.

How this Sensor Works

Each pixel on the GenX320 only sees motion. Either an increase in charge or a decrease in charge. Because of this, the camera can handle a very high dynamic range of lighting (however, the sensor bias settings may need to be tuned for specialized applications - you should contact prophesee for help if you need to do this).

By default, the sensor outputs 320x320 8-bits per pixel frames. A pixel value of 128 means that no motion has occurred. If a pixel senses an increase in charge then the value of the pixel will increase up to 255. Similarly, if a pixel senses a decrease in charge the value of the pixel will decreased down to 0. The amount of the increase or decrease is based on how many events happen with the set accumulation period. By default, the frame rate is set to 50 Hz which means each frame will contain information from events in a 20ms period. You can increase/descrese the framerate using sensor.set_framerate(). for example, by setting the frame rate to 200 Hz by doing sensor.set_framerate(200) will result in the camera output 320x320 frames at 200 Hz where each frame contains 5ms of events.

Finally, events themselves only add +1/-1 to the 8-bit bins. To increase the contrast of the image the events are multiplied by 16 by default so the image looks more vivid. You can change this multiplier via sensor.set_contrast() and you can change the default bin value from 128 by doing sensor.set_brightness(). For example, by doing sensor.set_contrast(64) the event image will become exceptionaly vivid at the expense of not being able to determine much anymore between a few and a lot of events per pixel. You can also apply a color palette by doing sensor.set_color_palette() to image.PALETTE_EVT_DARK or image.PALETTE_EVT_LIGHT for a more vivid looking image. However, note that this takes extra CPU time to apply the palette per frame and then even more CPU time to process frames. So, only use the color palettes for demo purposes.

Note that you can operate the camera in event mode if you recompile our firmware and turn off histogram mode. The processor will then generate a frame every time it receives 25600 events. In this mode the frame rate is high variable based on what's happening in the scene. However, this mode is highly inefficient for the processor and memory buses because the processor has to generate the frame through random access reads/writes to the frame buffer which means there will be very little CPU time left to do any actual image processing (which is why we do not enable this mode except through re-compiling our firmware).

Applications