IWR6843: Point cloud SNR differs depending on direction of motion

Alexander van Gessel

Part Number: IWR6843

I have been running some tests with the people counting lab, 3D ISK version, using the provided chirp configs, and I found that the radar has trouble detecting people walking away from the sensor. Compared to people walking towards the sensor, the point cloud has fewer points, and their SNR is lower, resulting in gtrack frequently failing to identify them.

This did not appear to be the case with the out-of-box demo. Is there some configuration that I am missing?

over 4 years ago

0 Alexander van Gessel over 4 years ago

Prodigy 40 points

Here are two screenshots from the python visualizer included in the lab.

The first image is of a person walking away from the radar, the second of that same person walking towards it.

0 Jackson Thomas over 4 years ago in reply to Alexander van Gessel

TI__Genius 14415 points

Hello,

Yes this issue is somewhat known. We don't currently have a workaround other than reducing the threshold values. You mention that this doesn't happen with the OOB demo, are you running the tracking algorithm or just looking at the pointcloud?

Regards,

Jackson

0 Alexander van Gessel over 4 years ago in reply to Jackson Thomas

Prodigy 40 points

With the OOB demo we are just using the IWR6843 to provide a point cloud that we cluster in an external program. That clustering algorithm currently ignores SNR, so it's possible that makes it more robust against this effect.

0 Jackson Thomas over 4 years ago in reply to Alexander van Gessel

TI__Genius 14415 points

Hi Alexander,

The 3D people counting demo also allows you to modify the SNR thresholds for track allocation, along with clustering requirements like number of points and consistent frames. See the following document for info on modifying these parameters.

3D_people_counting_tracker_layer_tuning_guide.pdf

Regards,

Jackson

Sensors

Sensors forum

IWR6843: Point cloud SNR differs depending on direction of motion