performance - Faster alternatives for calcOpticalFlowSF -
is there faster alternatives calcopticalflowsf? sooo slow , wanna run thing sequence of frames coming video. how can that?
there several methods optical flow based motion estimation have consider several things:
- are restricted cpu implementation / gpu's implementation decrease drastically run-time
- do need dense motion fields or set of sparse motion vectors / sparse of methods more scalable , need less run-time
- accuracy / high accuracy of dense methods critical on motion boundaries. in many application approximate dense motion field grid of sparse motion vectors , can use sparse methods pyramidal lucas kanade (opencv)
current libaries / methods are:
dense methods:
- opencv 2.4.4 provides on gpu broxopticalflow fast too
- the flowlib of gpu4vision group provides high accurate gpu implementation
- gpu implementation of tv-l1 on gpu provided
sparse methods:
- opencv since 2.4.2 provides pyramidal lucas kanade on gpu /earlier verions fast implementation on cpu
- the rloflib provides more accurate implementation gpu / cpu , matlab
- the gain adaptive lucas kanade / klt available gpu
you take current optical flow benchmarks, researcher provides link. common optical flow benchmarks middlebury , kitti
Comments
Post a Comment