IWR6843ISK: slow 3D object detection

Hi Shu-wei,

Performance expectation is the following:

1.  Frame time is 50 ms
2. Detection results from frame N are shipped at frame N+1
3. Visualizer takes up to 50 ms to parse and graph data (if it is significantly slower, you may need to put your PC in high performance mode.)

So if a person enters at time X:

1. 50 ms to complete the frame where they entered and get point cloud
2. 50 ms to output tracker data
3. 50 ms to graph

Giving 150 ms total. If the first frame where they are visible doesn't meet the detection thresholds, then add 50 ms. You can see that after 1 or 2 frames with weak detection, you get close to 300 ms.

If you are having issues with run-time performance, I recommend the following:

1. Start with the attached python script (it has .txt, but just change the file extension)
   7178.parserUseExample.txt

   import time
   from oob_parser import uartParserSDK
   
   parser = uartParserSDK(type ='3D People Counting') #3D People Counting labs (68xx)
   frameTime = 50/1000 #50 ms frame time
   try:
       uartCom = "COM" + str(19)
       dataCom = "COM" + str(18)
       parser.connectComPorts(uartCom, dataCom)
   except Exception as E:
       print(E)
       print('Com port connection failed')
   #open config file
   fname = 'mmw_pplcount_demo_default.cfg'
   cfg_file = open(fname, 'r')
   cfg = cfg_file.readlines()
   parser.sendCfg(cfg)
   
   while(1):
       print(parser.readAndParseUart())
       time.sleep(float(frameTime))  #50 ms

   1. This is used to parse the data from the device.
   2. You will need to modify the script to use the right points and the correct parser
   3. this will print data each frame
2. This will give lower latency - you can time the parser and try to increase its speed (or implement in C)
3. With a fast parser you can then do what you need to do at high speed with the data

Regards,

Justin

Thank you Justin. The advice is very helpful.