Long Short-Term Memory For Autonomous Driving Cars
Sprache des Titels:
Neural Information Processing Systems (NIPS 2016)
Currently much research on autonomous driving is based on inferring driving decisions from single-frame inputs without taking spatio-temporal correlations into account. We propose to exploit these correlations using a recurrent neural network (RNN) architecture known as Long-Short-Term Memory (LSTM) to make robust, more confident and more timely decisions. We applied current state of the art convolutional neural network architectures for semantic segmentation like FCN, ENet or SqueezeNet to image sequences. We identified three types of undesired artefacts and found that using the output of a segmenation-CNN as an input for a LSTM visibly improves the accuracy.