Space Occupancy Representation Based on A Bayesian Model for Unmanned Aerial Vehicles

Collision avoidance for unmanned aerial vehicles is a main task for autonomous aerial systems. One of the major challenges facing such technology is the existence of a high level of uncertainty in moving objects. In this research study, we chose to work with object’s classification and probabilistic models to deal with the system’s uncertainty. First, a classification takes place through detecting an object, identifying it using a trained convolutional neural network, and analyzing its velocity. Afterwards, a Bayesian probabilistic model takes inputs of the detected object’s type, orientation, and velocity along with the host unmanned aerial vehicle’s velocity. It gives an output of the detected object’s space occupancy. The simulation results show that the space occupancy clearly changes with respect to the available inputs of the Bayesian model as a function of time; hence, optimizing the space occupancy around the detected object.