Image Classification Using Sum-Product Networks for Autonomous Flight of Micro Aerial Vehicles

Flying autonomous micro aerial vehicles (MAVs) in indoor environments is still a challenging task, as MAVs are not capable of carrying heavy sensors as Lidar or RGD-B, and GPS signals are not reliable indoors. We investigate a strategy where image classification is used to guide a MAV; one of the main requirements then is to have a classifier that can produce results quickly during operation. The goal here is to explore the performance of Sum-Product Networks and Arithmetic Circuits as image classifiers, because these formalisms lead to deep probabilistic models that are tractable during operation. We have trained and tested our classifiers using the Libra toolkit and real images. We describe our approach and report the result of our experiments in the paper.