Open amiltonwong opened 6 years ago
I used RGB-D cameras, which means for each detected object in the image, you can crop it and compute x,y,z, from the depth point cloud. Each object detection is just a measurement in the pose graph.
Hi, @BeipengMu , After cropping bounding boxes in the depth image, how do you filter out background pixels that are too far away? Could you explain it detailedly? Thank you!
basically checking the depth values, background depth is usually much larger than foreground depth, check where the depth has jumps then break there. It's a simple heuristic, so the result could be noisy sometimes.
Hope this answers your question. Thanks!
On Tue, Jul 30, 2019 at 9:36 AM PYGUAN notifications@github.com wrote:
Hi, @BeipengMu https://github.com/BeipengMu , After cropping bounding boxes in the depth image, how do you filter out background pixels that are too far away? Could you explain it detailedly? Thank you!
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/BeipengMu/objectSLAM/issues/1?email_source=notifications&email_token=AEDD3K2BEKRY6XRR7KWHJOLQB6LI7A5CNFSM4FEEU63KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD3CPIYI#issuecomment-516224097, or mute the thread https://github.com/notifications/unsubscribe-auth/AEDD3K6Q3Z5CCRGQ2N7EB7LQB6LI7ANCNFSM4FEEU63A .
Hi, @BeipengMu , I use the Ostu method to select the depth threshold, but it seems to perform not well on TUM dataset. Could you give me some suggestions about the way to select the threshold? Thanks!
haven't tried Ostu method on TUM in particular. Usually these parameters are dataset and scene related
On Tue, Jul 30, 2019 at 10:15 AM PYGUAN notifications@github.com wrote:
Hi, @BeipengMu https://github.com/BeipengMu , I use the Ostu method to select the depth threshold, but it seems to perform not well on TUM dataset. Could you give me some suggestions about the way to select the threshold? Thanks!
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/BeipengMu/objectSLAM/issues/1?email_source=notifications&email_token=AEDD3K6JZYFIAKQIHTKEBMDQB6P4JA5CNFSM4FEEU63KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD3CRC7A#issuecomment-516231548, or mute the thread https://github.com/notifications/unsubscribe-auth/AEDD3KYOLO2N6AHVQO2OR6LQB6P4JANCNFSM4FEEU63A .
Hi, @BeipengMu ,
Could you list the steps on how to apply RCNN detection with objectSLAM?
Thanks!