Closed plushpluto closed 4 years ago
It depends on the accuracy of depth prediction in the testing scene, regardless of the platform on which the camera is mounted (in your case, UAVs). You would need a fairly powerful machine if you plan to do onboard computation. I tested CNN-SVO on an Nvidia Xavier on a mobile robot and it worked. There are two main considerations, though: (1) the memory limitation (which means that you cannot build a big map), and (2) the processing frame-rate (which means that you cannot move the robot quickly).
Is this suitable for UAVs?