Nvdsanalytics Plugin Line Count error

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi
I have installed a jetpack 4.4 version in a jetson nano, then downloaded deepstream as a tar file and installed it.
While executing the deepstream nvdsanalytics-test, to tryout the line-count, i got some problem:

  1. I have drawn a line based on the coordinates of line-crossing 980;840;980;300;980;540;1750;540 and gave a input video (highway traffic video - https://drive.google.com/open?id=145z5bwcAVIBZZkAmDw4ApwOS9Lq1yrji) in the video there are 39 cars by visually counting…,but the LC-Count shows only 27.

  2. Then i experimented same line-counting to count persons in a video, but it did’nt count the persons.
    (https://drive.google.com/file/d/1hGZdkGIq8CDgwWFintSABfo5kz4jtJwh)

Note: I have also changed the class-id in the config file ,but nothing appears to be changed.The person’s crossing the line did’nt got counted.
(Did anyone face this type of problem)

  1. The overcrowding count also had the same problem. Not even a single person did’nt get counted while the cars were getting counted.

My Question is,

I solved the frame dropout issues based on the link below.

(The deepstream-nvdsanalytics-test frame dropped problem is very serious - #4 by CoderJustin)

Based on the interval value the count increases to the correct value(39).When i give the interval value as 2 the lc-count shows 27 but at interval of 3 lc-count shows 26.

  1. What does a interval mean here?

  2. How does the line count takes the value while vehicle crosses it? & Is it class dependent, if yes how to change it to the desired class?

3 . What is cumulative count, How its been calculated ?

2

I’ve met with the same person counting problem, turns out I forgot to change classid to 2, finally works well. Please make sure your direction is correct, and the target is complete crossing the line.
In my opinion, the line crossing couting depends on the tracking result, for example, if the tracking label of a person is ‘person 1’ before he or she crossing the line, but somehow the neural net lost tracking when this target crossing the line, and track back the target again after line crossing, but the tracking id is ‘person2’ now, then it won’t be counted, since the id is different, person1 didn’t show at the other side of the line.
Just my thought, not quiet sure.