If my algorithm detects the type of object, how should I know if that object is moving or not? Suppose a person carrying an umbrella. How to know that the umbrella is moving?

I am working on a project where I want to know whether that particular object belongs to the person entering inside the store. I was thinking about the bounding boxes(bb) approach where if the person's bb overlaps with the object's bb. But the problem arises when there are multiple objects with a person.

I would appreciate your help. Thanks.

  • $\begingroup$ Your title ask for moving object but your post ask about if an object belong to a person. What is your question ? $\endgroup$ – Jérémy Blain Nov 28 '18 at 9:51
  • $\begingroup$ sorry for the confusing sentence. If the object moves with a person then it belongs to the person. This is the logic I want to use. I will check the start, middle and final frames to get the results. $\endgroup$ – prb Nov 28 '18 at 9:55
  • $\begingroup$ I have tried the overlap between the bounding boxes approach. If the intesection over union between two bounding boxes is large then object belongs to the person . again I will check the frame for that $\endgroup$ – prb Nov 28 '18 at 9:58

Either (a) re-recognize it in subsequent frames and then train a network employing a trajectory model to the changing object model parameters or (b) recognize the object and its motion in a single object-trajectory parameterized model from the sequence of sets of detected edges indicating edge movement in the sequence of frames.


You can use several consecutive frames as an input. If you have dataset of moving/not moving objects, you can use it to train your model. You can use CNNs with additional time dimension in this case, for example.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.