Object detection in H264 Videos

asked 2019-10-15 02:15:31 -0500

Neelam Jain gravatar image

Hello, I want to track object in videos. I understand there are multiple examples. However, it seems all examples provides a rectangular coordinated. I would like to get the exact coordinates for any object. For e.g. my application needs to replace a ball with another object from different video or i may want to change the color of cloths

edit retag flag offensive close merge delete


it sounds like Mask R-CNN will be your best option. If you eventually follow the example, the bounding box coordinates are found as (x, y) tuples as explained in this Github issue.

Erik Langskjegg gravatar imageErik Langskjegg ( 2019-10-15 03:36:07 -0500 )edit

There is also a nice tutorial for Mask R-CNN and OpenCV on PyImageSearch

Erik Langskjegg gravatar imageErik Langskjegg ( 2019-10-15 03:45:56 -0500 )edit

Mask rccn is pretty expensive - computational spoken. Maybe a "normal" cnn would do the job too. It depends also on the target runtime which network(architecture) to use.

holger gravatar imageholger ( 2019-10-15 15:30:48 -0500 )edit