Administrator
|
There is no option for multithreading in RTAB-Map. Though the GUI, the core and the camera run on 3 different threads. The underlaying algorithms (features detection/extraction/matching) depend on OpenCV methods like those in Feature2D module and FLANN kd-tree. I don't know if OpenCV can be built to use OpenCL or TBB on these methods. The Bayes filter and likelihood computation could be multithreaded though. For other algorithms used, TORO or g2o are used for graph optimization.
If you have CUDA, you could build OpenCV with GPU support and select the GPU version of the features in RTAB-Map (like SURF-GPU and ORB-GPU).
Increasing RTAB-Map speed could increase the maximum size of the Working Memory usable in real-time (under the real-time threshold), so a bigger local map that can be kept and updated online.
|