Abstract

The aim is to identify the road users that the driver is observing at an instance in time. This is achieved by merging his vision parameters with the environment model of the vehicle. Field studies are conducted with a head-mounted eye tracker to obtain the 3D gaze coordinates of the driver. A pipeline is created for automatic label generation from these drives and to quantify the accuracy of this generated dataset. One key contribution is that this method removes the requirement of human annotators. Finally, different neural network designs are evaluated for their performance on this generated multiclass, multilabel dataset and on different traffic densities.

Results/Implementation/Project Description

Conclusion

[ PDF (optional) ] 

[ Slides Kickoff/Final (optional)]