AWR1843BOOST: Lab0008 Automated Parking: sometimes two clusters appears for one real object in different regions of the elevation plane

Hello,

First issue

Is the epsilon of DBscan affects the size of cluster in GUI? I am debugging the firmware and I found the epsilon around 0,40006 and minPts is 3. visualizer sometimes detects a chair in particular range from radar as two clusters. I have increased epsilon to 0.6, I found change in cluster size with the same scene, I need you to verify my experiment whether it's correct or not?

Second issue

sometimes chair are not detected when I put two chairs on front of Radar within 1:2 meters

Third issue

Q2. Is parking Automated tested well in indoor environment? I the result isn't reliable, I am simulating parking area in the office, I put the Radar on a small table with wheels and move it toward the objects like chair, fan, office, my results was Automated parking application is detectable to wall, if the there more than object in the scene within 1:3 meter results becomes ambiguous and I can't identify all objects inside the visualizer!

I have another question

Q1. What's the default threshold of CFAR  number in Automated Parking project and how to change it from firmware?

Q2. In the following image I put the radar on 60 cm away from the table but visualizer didn't show a red cluster or even detected points within 1 meter. I moved the radar forward and back but didn't show any cluster related to the table.

Thanks,

Hi,

Before we discuss the Clustering Issues, let's discuss the detection issues (point cloud)

Second Issue

One thing that can help to understand the limitations of an environement is to run the mmWave SDK OOB demo. The SDK OOB demo visualizer provides the "Range Profile for zero doppler" which allows to see the static clutter in the scene. Usually if you are not able to clearly distinguish the peak of the chairs in the profile you will not be able to detect them statically. If you move the chairs, then you can see in the point cloud if a detection is present. Moving object is easier to detect than static.

Third Issue

The Automated Parking processing chain was tested outdoors. Inside a closed environment there is additional clutter and it is sometime hard to detect.

When there are several objects if the objects are static, in the same range bin, then only one will be detected. If the objects are moving at the same speed than there could also be detection challenges.

For the test with the table, in order to better understand the environment it would help to run the oob demo and check the "Range Profile for zero doppler"

Thank you

Cesar

Hello Cesar, 

I tested the same scene in the above image (Table on front of the radar within 60 cm) with OOB demo and gave the same result (No detection of the table)

When I am tuning the CFAR range threshold, table gets detected (one detected points appears on 60 cm).

Q1. Is CFAR range threshold affects on the objects detection?

Q2. What's difference between CFAR range threshold and Doppler range threshold and what's the affect of each at general? 

Q3. How to achieve zero doppler in OOB demo?

Thanks,

Hello Cesar,

Thanks for you answer

I found this video from TI on youtube www.youtube.com/watch

But tested on Lab0007 MRR project not Automated Parking

Could you provide me with a test video for automated parking, with test case of a multiple objects on front of radar starting from steady state then moving towards the objects ?

Thanks,

Mostafa