Imaginghub Forum - Machine learning model deployed on my drone

A couple of my friends and I have developed a machine learning model for object detection. At inference time it should use images captured from the drone. In other words, it should detect objects on the views available from the heights. And a drone should be used to take such images. This should be a real-time (close to) prediction. But what we are worried about, is the processor installed on our drone is capable to deal with all needed calculations for making a prediction. What other possible solution to the problem can you suggest?

Answers

Object detection task is a resource-consuming task. I think your computer vision model is complex enough. Object detection is available with deep neural networks which indeed require a large number of computations. Am I right that you have used deep learning algorithms when creating your model?

Yes, absolutely. We weren't able to achieve any decent results without DL.

Then I suppose your microprocessor installed on the drone is not powerful enough to do all the things in real-time. At the moment, I see only 2 possible ways. First is to install several more processors on the drone and amend the model such that is can split the computations between several processors (perform parallelization). Another solution is possible if your drone has the ability to connect to the Internet through a mobile network (and if it operates in the region where the mobile network for the Internet is available). You can host the model somewhere in the cloud and then access it directly from the drone. The drone will not need to do any computations, it should only send images to the cloud, and then receive a response from it. But, once again, it is possible only if you have a good mobile data coverage in the region of operation. Also, be aware of the costs of mobile Internet connection and costs for using cloud.

Thank you for your answer. It seems that we had to think about this problem beforehand.

Add Answer

Machine learning model deployed on my drone

Machine learning model deployed on my drone

Answers

Need support?