Real-time 3D Particle Tracking Using Dynamic Vision Sensors
David Borer, Thomas Rösgen
General description
Flow visualization in wind tunnel testing can provide an intuitive insight into flow phenomena. The simplicity and responsiveness of the visualization method is of crucial importance for practical use. In this work we aim to develop a cost-effective and fast visualization method based on tracking neutrally buoyant soap bubbles with a set of high speed motion-sensitive cameras. The system is applicable to large scales (on the order of meters) and provides continuous time resolved 3D particle tracks. By seeding the region of interest with suitable tracers the flow field can be visualized and information on the prevailing flow features can be derived.
Dynamic Vision Sensor
The "Dynamic Vision Sensor" (DVS) registers the temporal change in illumination with very low latency, capturing fast processes while generating only low data rates. In contrast to frame based cameras, the pixels of the Dynamic Vision Sensor operate individually. The pixels are sensitive to the time derivative of the logarithm of the intensity; this design offers a high dynamic range of reportedly 120dB. Intensity changes are reported as a stream of events; an event is the basic unit of information consisting of the pixel address, the time instant of intensity change and its sign. The operating principle offers an inherent staionary background suppression and reduces the data volume to be processed considerably. The temporal resolution of 1 microsecond enables an accurate registration of fast moving objects without the need for conventional high speed cameras.
More information on the DVS camera is available at external pageiniLabscall_made.
In this work both the 1st and 2nd generation of the DVS camera have been used. The 2nd generation called the Dynamic and Active-pixel Vision Sensor (DAViS) offers a higher resolution and can additionally capture frame based images.
Tracking
The 3D reconstruction of the particle tracks requires a set of calibrated cameras, i.e. one has to know how a point in the 3D environment
relates to a projected point on the image sensor. This relation is established prior to the measurement with a calibration target using a known pattern of blinking LEDs. Two cameras already provide an operable 3D tracking system; a higher number of cameras can be used to increase the probability of detection in areas of poor lighting, interfering reflections or poor background rejectability.
The particle tracking scheme employs a Kalman filter to estimate the particles' position and velocity. The Kalman filter offers an ideal
framework to process the sequential data from the Dynamic Vision Sensors and it can handle 2D tracking and 3D reconstruction in a similar manner. The 3D tracking algorithm combines the data from multiple cameras in a straightforward fashion; the number of cameras is not limited. The proof-of-concept installation in the IFD wind tunnel works with three cameras.
Example Case: Jet in Crossflow
To demonstrate the measurement technique it is applied to the flow field produced by a jet in crossflow.
The measurement was performed in the wind tunnel of IFD. The jet is produced by a blower connected to a nozzle (diameter 0.12 m) which is positioned below the wind tunnel floor. The wind tunnel provides the crossflow at a speed of 3.2 m/s. The data recorded by the DAViS camera is shown in the following video.
The setup consisted of 2 cameras covering a volume of slightly less than 1 cubic meter. The resulting particle tracks were interpolated onto a regular grid in order to calculate streamlines. A comparison between a subset of raw particle tracks and the interpolated streamlines is shown in the following video.