Design of an Artificial Neural Network to Detect Obstacles on Highways through the Flight of an UAV

Due to several risks that involve travelling in highways, such as vehicle collision, natural disasters and other obstacles, the need of implementing an early warning system which detects obstacles to provide security has been growing in the last years. Artificial neural networks based systems have been successfully applied in obstacle detection through image processing and recognition. To address this issue, a Multilayer Feed-forward Network (MFN) was designed to detect obstacles in highways from a zenith perspective. Backpropagation algorithm is used to supervise the training of the proposed neural network by minimizing the square error function via descending gradient criteria. Performance analysis was performed by using binarized, grayscale and RGB images with constant size and considering several contexts. The proposed neural network was trained and tested offline. In a next stage of the project, this neural network will be implemented into the on-board computer of unmanned aerial vehicles (UAV’s) as an early warning platform.