# Prophesee EVK4

Introduce EVK4, point to doc, setup example by changing bias etc...
include use with kinect.
recording datas. getting right format.

## Introduction and setup

The Prophesee EVK4 is a dynamic vision sensor that delivers events when a change of brightness occurs. More precisely, ON and OFF events are generated when the brigthness rises or lowers respectively. It can generate events on a scale of nanoseconds and has a resolution of 1280x720. The installation and setup can be found on the Prophesee [website](https://docs.prophesee.ai/stable/get_started/index.html)

In our case, we will use a robotic setup where the camera is mounted on top of a table with a robotic arm in sight. The purpose is to observe objects moving and combine the events generated by the EVK4 with a Kinect depth camera to perform sensor fusion.

## Tuning Bias

The most important parameters to consider in order to calibrate the camera to your environment are the [bias](https://docs.prophesee.ai/stable/hw/manuals/biases.html#chapter-biases). A well calibrated camera can make a tremondous difference in the quality of the datas generated. As examples, we recorded a ball rolling on the table and displayed the events at 100fps with an accumulation time of 20ms. This sample is a raw recording without any tuning.

![test](https://github.com/rouzinho/Neuromorphic-Computing/blob/main/img/no_bias.gif?raw=true)

In this one, we changed the bias settings.

![test](https://github.com/rouzinho/Neuromorphic-Computing/blob/main/img/bias.gif?raw=true)

## Event Processing

The details of the following parameters can be found [here](https://docs.prophesee.ai/stable/hw/manuals/esp.html#event-trail-filter-stc-trail)

In case of noisy lightning with direct sunlight, the anti-flickering parameter can perform some filtering. In our setup, this parameter was not necessary however you can have an insight of what the events should look like. 
![test](https://github.com/rouzinho/Neuromorphic-Computing/blob/main/img/anti_flicker.gif?raw=true)

The more useful parameter is the event trail settings. Depending on the light source conditions, a moving object always leaves a certain trail of events. We tuned here the trail filtering with a CUT option. 

![test](https://github.com/rouzinho/Neuromorphic-Computing/blob/main/img/cut_trail.gif?raw=true)

The KEEP trail option is also available>

![test](https://github.com/rouzinho/Neuromorphic-Computing/blob/main/img/keep_trail.gif?raw=true)

## ROI

Finally, it is often necessary to select a Region of Interest for a more efficient generation of events. It allows to remove unnecessary noise and focus on a specific location of the scene.

![test](https://github.com/rouzinho/Neuromorphic-Computing/blob/main/img/roi_up.gif?raw=true)

Same ROi but with the object moving horizontally to demonstrate the resized window of events.

![test](https://github.com/rouzinho/Neuromorphic-Computing/blob/main/img/roi_left.gif?raw=true)