Closed liquidcronos closed 5 years ago
A first approach was found using the sum of relative positions, this means that things moving in the same direction for an extendet period of time are penalized in favor of oscilating random movements.
Added new approach which estimates immobile objects by imposing constraints on the OF. One should be careful with this however, because it allows for small moving objects to be incorrectly categorized as moving. a combinations of position and velocity constraints could solve this problem
Using constraints on OF and distance from the first position accurately detects independently moving objects. The next step would be to use the IMU or State data of the Drone to dynamically prescribe these restraints (maybe even for each point separately)
the Distance from the starting point is a poor measure of moving targets as it discards a lot of useful information inherent in the image. There aim is to find a better classification using the regular patterns moving objects make contrasted with the pseudorandom pattern the immobile features generate.