Note
This Python sample is available with all Metavision Intelligence Plans. The corresponding C++ sample is available only with our Professional plan.
Particle Size Measurement using Python
The Analytics API provides algorithms to both count and estimate the size of fast moving objects.
The sample metavision_psm.py
shows how to use the python bindings of Metavision Analytics SDK to count,
estimate the size and display the objects passing in front of the camera.
We expect objects to move from top to bottom, as in free-fall. But, command-line arguments allow to rotate the camera 90 degrees clockwise in case of particles moving horizontally in FOV or to specify if objects are going upwards instead of downwards. An object is detected when it crosses one of the horizontal lines (by default, 6 lines spaced 20 pixels apart are used). The number of lines and their positions can be specified using command-line arguments. Detections over the lines are then combined together to get a size estimate when the object has crossed the last line.
Expected Output
Metavision Particle Size Measurement sample visualizes the events (from moving objects), the lines on which objects are counted, the detected particles, their estimated sizes and the total object counter:
Setup & requirements
To accurately count and estimate the size of objects, it is very important to fulfill some conditions:
the camera should be static and the object in focus
there should be good contrast between the background and the objects (using a uniform backlight helps to get good results)
set the camera to have minimal background noise (for example, remove flickering lights)
the events triggered by an object passing in front of the camera should be clustered as much as possible (i.e. no holes in the objects to avoid multiple detections)
Also, we recommend to find the right objective/optics and the right distance to objects, so that an object size seen by the camera is at least 5 pixels. This, together with your chosen optics, will define the minimum size of the objects you can count.
Finally, depending on the speed of your objects (especially for high-speed objects), you might have to tune the sensor biases to get better data (make the sensor faster and/or less or more sensitive).
How to start
To start the sample based on the live stream from your camera, run:
Linux
python3 metavision_psm.py
Windows
python metavision_psm.py
To start the sample based on recorded data, provide the full path to a RAW file (here, we use a file from our Sample Recordings):
Linux
python3 metavision_psm.py -i 195_falling_particles.raw
Windows
python metavision_psm.py -i 195_falling_particles.raw
To check for additional options:
Linux
python3 metavision_psm.py -h
Windows
python metavision_psm.py -h
Code Overview
Pipeline
Metavision Particle Size Measurement sample implements the following pipeline:

Optional Pre-Processing Filters/Algorithms
To improve the quality of initial data, some pre-processing filters can be applied upstream of the algorithm:
metavision_sdk_core.PolarityFilterAlgorithm
is used to select only one polarity to count the objects. Using only one polarity allows to have the sharpest shapes possible and prevents multiple counts for the same object.metavision_sdk_cv.TransposeEventsAlgorithm
allows to change the orientation of the events. Note thatmetavision_sdk_analytics.PsmAlgorithm
requires objects to move from top to bottom, and if your setup doesn’t allow it, then this filter/algorithm is useful for changing the orientation of events.metavision_sdk_cv.ActivityNoiseFilterAlgorithm
aims to reduce noise in the events stream that could produce false counts.
Note
These filters are optional: experiment with your setup to get the best results.
Particle Size Measurement Algorithm
This is the main algorithm in this sample. The algorithm is configured to count objects while they’re passing from top to bottom in front of the camera.
To create an instance of metavision_sdk_analytics.PsmAlgorithm
, we first need to gather some configuration
information, such as the approximate size of the objects to count, their speed and their distance from the camera,
to find the right algorithm parameters.
Even though a specific accumulation time might provide a well-defined 2D shape, the algorithm doesn’t directly have access to it because it’s only seeing events through the lines of interest. Ideally, we would like the object to advance one pixel at a time, so that in the end the algorithm could process all the information in this well-defined 2D form. This is however not compatible with a large accumulation time. The accumulation time must therefore be decoupled from the time interval between two line processings. That’s why we have two temporal parameters:
Precision time: Time interval between two line processings. Should roughly be equal to the inverse of the object speed (pix/us)
Accumulation time: Temporal size of the event-buffers accumulated on the line for a line processing. Should be large enough so that the object shape is well-defined
The lines should be close enough so that there’s no ambiguity when matching together detections done on several lines.
Once we have a valid calibration, we can create an instance of metavision_sdk_analytics.PsmAlgorithm
.
The metavision_sdk_analytics.PsmAlgorithm
relies on the use of lines of interest to count and estimate the size of the objects
passing in front of the camera and produces metavision_sdk_analytics.LineParticleTrackingOutput
and metavision_sdk_analytics.LineClusterWithId
as output. A metavision_sdk_analytics.LineParticleTrackingOutput
contains a global counter and object sizes as well as their trajectories.
The global counter is incremented when an object has been successfully tracked over several lines of interest.
The algorithm is implemented in an asynchronous way which allows to retrieve new estimations at a fixed refresh rate rather than getting them for each processed buffer of events. Not only is this more efficient for visualization purpose but it’s also easier to process results only when a new particle has been detected.
Frame Generation
At this step, we generate an image that will be displayed when the sample is running. In this frame are displayed:
the events
the lines of interest used by the algorithm
the global counter
the reconstructed object contours and the estimated sizes
The metavision_sdk_core.OnDemandFrameGenerationAlgorithm
class allows to buffer input events
(i.e. metavision_sdk_core.OnDemandFrameGenerationAlgorithm.process_events()
) and generate an image on demand
(i.e. metavision_sdk_core.OnDemandFrameGenerationAlgorithm.generate()
).
Once the event image has been generated, following helpers are called to add overlays:
metavision_sdk_analytics.CountingDrawingHelper
: timestamp, global count and linesmetavision_sdk_analytics.LineParticleTrackDrawingHelper
: object sizes and contoursmetavision_sdk_analytics.LineClusterDrawingHelper
: clustered events along the lines
As the output images are generated at the same frequency as the metavision_sdk_analytics.LineParticleTrackingOutput
produced by the
metavision_sdk_analytics.PsmAlgorithm
, the image generation is done in the metavision_sdk_analytics.PsmAlgorithm
’s output callback:
def psm_cb(ts, tracks, line_clusters):
events_frame_gen_algo.generate(ts, output_img)
counting_drawing_helper.draw(ts=ts, count=tracks.global_counter, image=output_img)
detection_drawing_helper.draw(image=output_img, line_clusters=line_clusters)
tracking_drawing_helper.draw(ts=ts, image=output_img, tracks=tracks)
particle_sizes = []
for track in tracks:
particle_sizes.append("{:.1f}".format(track.particle_size))
if particle_sizes:
print(f"At {ts}, the counter is {tracks.global_counter}. New particle sizes (in pix): ["
+ ", ".join(particle_sizes) + "]")
window.show_async(output_img)
if args.out_video:
video_writer.write(output_img)
The event processing is done while iterating over metavision_core.event_io.EventsIterator
:
# Process events
for evs in mv_iterator:
# Dispatch system events to the window
EventLoop.poll_and_dispatch()
# Process events
if filtering_algorithms:
filtering_algorithms[0].process_events(evs, events_buf)
for filter in filtering_algorithms[1:]:
filter.process_events_(events_buf)
events_frame_gen_algo.process_events(events_buf)
psm_algo.process_events(events_buf)
else:
events_frame_gen_algo.process_events(evs)
psm_algo.process_events(evs)
if window.should_close():
break
Display
Finally, the generated frame is displayed on the screen. The following image shows an example of output:
