Associate clusters accross frames for tracking
Hello everyone! I'm currently working on a project, where I'm trying to track moving objects with a fixed, non moving Lidar Scanner. I applied several filters to my PointCloud Data and applied an euclidian cluster extraction, based on this Github Code: Cluster Extraction Code
For the Bounding Boxes, I used getMinMax3D to compute the cluster-centers and draw a cube around each cluster.
Now I'm having trouble chosing the right way to move forward. The idea is to somehow associate the clusters between one time step and the next. The most commonly referenced method for the actual tracking is a Kalman filter with a constant velocity model.
However, I'm not sure which functions/resources there are, to associate the clusters, which are newly generated in each frame. My research only yielded options like the hungarian algorithm, to associate Clusters accross frames, however I didn't find any helpful implementations. For the filtering and extraction I was able to rely on the many available tutorials and examples, which isn't the case now.
I would be happy about any suggestion. Thank You in Advance
Asked by th6262 on 2018-08-07 01:41:06 UTC
Comments
Hey I have the same problem, I have bounding boxes of detected objects but there are some false positive, size errors, moving errors, etc. So now I need to track them in time and smooth the movement and size of the bounding boxes. Did you come up with a solution?
Asked by autonomobil on 2019-04-25 10:36:04 UTC
Hey, take a look on https://github.com/tue-robotics/wire
Paper: https://pure.tue.nl/ws/portalfiles/portal/3390005/22076505621656.pdf
Asked by r7vme on 2019-05-11 12:27:47 UTC
Thanks, I had a look a that, very helpful.
Asked by autonomobil on 2019-09-03 07:47:37 UTC