AWR1642: Frames required for a stable detection

Hi Anil,

We have already tried the suggestions provided in the thread which you directed me to. Also about your suggestion regarding a video, I am attaching the following video. In this video the configurations are as I described before. The .bin file we used to here is the original mmw-demo .bin file.

You can observe a bunch of things in this video which depict false positives , too many points under the bridge , points popping up inside our ROI ( Shown using red parallel lines on the visualizer ). I am sorry about the rainy scenario in the video as we recorded it today ans its been raining here.

Looking forward to your inputs based on our observations. 

Thanks

Hi Amol,

Thanks for the video. Could you try another capture with the configuration that I've provided below? I have increased the number of chirps per frame (from 16 to 128), which should improve the SNR.  I have increased the SNR threshold for CFAR from its previous value (about 15 dB)) to about 18 dB. I have also disabled the multi-peak option, and removed the MIMO configuration. Each of these options increase the SNR, and reduce the possibility of false detections.

Since TI's EVM has an elevation FoV of about 10 degrees, objects above and below the radar will be detected. For an ACC application, you should redesign the antenna to have much lower elevation FoV ( say 1 degree or lesser). 

Modified configuration begins below :   

sensorStop

flushCfg

dfeDataOutputMode 1

channelCfg 15 3 0

adcCfg 2 1

adcbufCfg 0 0 1 1

profileCfg 0 77 2 5.30 52.68 0 0 15 1 256 6250 0 0 40

chirpCfg 0 0 0 0 0 0 0 1

frameCfg 0 0 128 0 50 1 0

guiMonitor 1 0 0 0 0 0

cfarCfg 0 0 8 4 4 0 6144

cfarCfg 1 0 8 4 4 0 6144

peakGrouping 1 1 1 1 224

multiObjBeamForming 0 0.5

sensorStart

multiObjBeamForming 0 0.5

Regards,

Anil

Hi Anil,

We implemented the configuration which you have recommended over here. Although your recommendation helped to reduce noisy detection it is still not sufficient to enable better detection. in the following video you can observe that there are still noisy points popping up in the ROI which is a big obstruction for our desired ACC application. Could you please suggest more methods to overcome this issue? We are also trying some tweaks on the configuration and would let you know if we get results better than what we have right now.

Thanks 

Hi Amol. 

I watched the video, and I don't see spurious detections. Every object detected seems to correspond to an object in the scene. Could you freeze a frame with the problematic detection and publish it? 

Also, I noticed that you've set the range depth to about 10 meters, even though the configuration allows for upto 50m. Is there a reason for this? 

Regards,

Anil

Hi Anil,


Thank you for watching the video. I think if you closely watch the video again, there are some noisy points that you can observe. For instance notice at 0:34 that despite the vehicles being on the left side you can still see some objects being detected in the center when there were o objects at all, also 1:10 onwards, you can see that there are points in the very close range (<1 m) which are visible on the visualizer.

One more thing, the range depth is at 50m in the configuration. How did you think that it was 10? The visualizer shows the arc only till 10m always.

Also finally, we are transmitting the detected objects in to our C code by using a javascript file that finds them on the visualizer and sends them to the C code using socket io , when we do this we can see points in close range ( 2-7 m) inside the ROI ( 1m on both left ad right from the center) whenever oncoming cars from the left pass though our car. This makes our ACC system think that there is a car in the center lane and that lowers the efficiency of our application. Could you suggest some modifications for this ? I am willing to provide you more information/ multimedia if required.

Also in response to your configurations, we tried increasing the SNR ratio which did not really make much difference on the detections. However, when we increased the threshold values more ( 10000) we were able to eliminate spurious detections but that also led to loss of regular detections. 

Thanks.