This is a CUDA accelerated version of the ORB-SLAM2. OpenCV CUDA library, custom CUDA kernels and pipeline execution are used to accelerate the bottleneck functions.
The algorithm is tested with the EuRoC dataset and the KITTI dataset. The test results show that a 2x end-to-end speedup is achieved.
This is a course project of CSC2231 at University of Toronto supervised by Professor Nandita Vijaykumar.