DeepSignals: Predicting Intent of Drivers Through Visual Signals

June 2019

tl;dr: Detecting automotive signals with CNN+LSTM to decode driver intention.

Overall impression

Vehicle light detection is a rather overlooked field in autonomous driving, perhaps due to the lack of public datasets. As long as autonomous cars and human driver co-exist, the capability to decode human driver’s intention through visual signal is important, for vehicle-to-vehicle communication.

The paper’s performance is not that good. Perhaps due to the severe imbalance in the dataset.

Key ideas

Technical details