Processing time differs when used ros topic as input for disparity computation

asked 2017-05-03 04:04:09 -0500

mkreddy477 gravatar image

updated 2017-06-28 06:49:55 -0500

Hello All,

I have observed performance drop while using using subscribed ROS topic as an input to that of the processing over the video frame.

To investigate further I have followed the following steps:

1) Initially I have used video recorded from a bag file as Input to the standalone stereoBm disparity algorithm. The compression technique used here is MJPEG.

2) To clarify the fastness might be due to the compression technique used, I have recorded video without any compression. I have found same execution time for both the videos.

3) Also, I tried using the same videos as inputs inside the ros framework. I haven't observed the performance gain for the algorithm.

Can anybody explain me why there is around 2x performance drop observed for the stereo BM algorithm in ROS frame work.?

Edit1: I have done some more analysis. I have observed the performance of stereo BM saturated at some point. I mean I have customized the BM algorithm to use multiple threads. Till 8 threads I have seen comparable speed with the stand alone one. After 8 threads the computational time for the algorithm is saturated in ROS.

One more observation is that the time taken for the computation is decreased if the bagfile is played at 2.0 rate.

Can any one help me in debugging the issue.

Thanks and regards,

Murali krishna

edit retag flag offensive close merge delete