rpg_svo
rpg_svo copied to clipboard
uzh-rpg
Doesn't seem to create new keyframes...
Hello, I'm trying to get SVO to work for my robot, but I'm having trouble with it.
I'm using a PSEye camera (640x480 / 60fps), wider angle setting, which I calibrated using ethzasl code as recommended to an error ~ 0.3.
It looks like it first identifies lots of good feature points, and tracks them successfully, but as they move out of view, new feature points are not identified/tracked, and I hypothesized this was because for some reason new keyframes were not being created (and the initial keyframe was moving out of view). I am pretty sure I'm moving it more than 15% depth value which is supposed to trigger a new keyframe. I've attached a sequence of images showing this behavior. What do you think is going on? Could it be something as simple as bad/misspecified calibration?
Thanks for your help!
I find this problem, too. It seems that this package only detect features at start only.
you are probably rotating camera without any translation this kind of movement is a killer for most monocular slam systems. try to shift camera instead of rotating it. also u can try to change GridSize in Config class to smth smaller like 20 or 15
@inspirit，I got this problem too，and when I just moved the camera away from origin place a little bit(just 1 meter far)，then the SVO fail.......
Hello! After many tests, I found that, this problem arises when features are not enough. So, I try to make the environment in a mess, and svo runs well. Although this ' solution ' may seems not a real solution, I don't have other methods now. Maybe we should modify svo itself to solve the problem totally.
with best wishes!
An Liang from Tsinghua Univ. of China 在 2016-02-03 19:34:38，"alejodosr" [email protected] 写道：
Same problem here.. any found solution?
— Reply to this email directly or view it on GitHub.
The initialization is very important, and also a good camera. I got a close result as the data set, using a ov7251 camera.
