Object tracking - Githubissues

ceccocats / tkDNN

Deep neural network library and toolkit to do high performace inference on NVIDIA jetson platforms

GNU General Public License v2.0

718 stars 209 forks source link

Object tracking #163

Closed alex-stonebridge closed 3 years ago

alex-stonebridge commented 3 years ago

Hi, firstly, I want to say thanks for this repository. I have my custom Yolo NN working perfectly fine with a Xavier NX. As a next step I would like to implement object tracking. Could you probably share some ideas how to add this feature having your architecture as a base?

mive93 commented 3 years ago

Hi @alex-stonebridge There are many option, here's what we implemented in our projects:

Multi tracker Extended Kalman filter on a single point representing the object (as done in this project https://github.com/mive93/class-edge)
Feature based tracking, computing the features of each BBs with classical methods (e.g. Harris, FAST etc) and track them frame by frame
tkDNN Center track implementation, in this branch https://github.com/ceccocats/tkDNN/tree/cnet