Performances improvement of back propagation algorithm applied to a lane following system

In this paper, a new high-level hardware FPGA design methodology, for artificial neural networks (ANN) descriptions, is proposed. In order to speed convergence of an ANN while avoiding instability, a back-propagation algorithm (BP), employing a momentum term to train the ANN, is used. A case study of comparison, between the back propagation (BP) and back-propagation with momentum (BPM) algorithms, is proposed. Matlab is used to validate this comparison. To achieve our goal, the two proposed design algorithms are implemented using Altera Cyclone FPGA. The originality of the work resides in the experimental validation of the simulation results with a car like robot using a lane following system. Numbers of parameters like training iterations, neurons in the hidden layer are used during this analysis. The simulation results show that the ANN trained by BPM algorithm will provide better performances than that trained by simple BP algorithm. The hardware implementation using Altera chip FPGA shows that BPM algorithm consumes less resources than the standard BP algorithm.