Detection of VRUs using Deep Learning

Detection of Vulnerable Road Users (VRUs), including pedestrians and cyclists, using Deep Learning techniques and computer vision. Experiments were carried out at the Campus of the University of Alcalá (UAH) by the INVETT Research Group using the DRIVERTIVE vehicle.




Detection of Drivable Area

Deep-Dig performs road segmentation and detection of drivable space for autonomous vehicles and driver assistance systems. It is based on Deep Learning techniques using a powerfull ResNet101 Convolutional Neural Network model together with Advanced Data Training Augmentation techniques.




Blind Sport Warning

BSW issues warnings to drivers whenever a vehicle is detected in the blind spot region of the driver's mirrors. It works at daytime and nighttime in all wheather conditions. Operation is robust in all types of environments including tunnels and roundabouts. The system provides accurate detection of cars, trucks, motorcycles and bikes using a single camera.




Pedestrian Detection and Protection

PDP is a stereo-based system for pedestrian detection and tracking. Pedestrians in the surrounding of the vehicle are analysed and tracked, with special focus on those pedestrians located in the vehicle trajectory or on the edge to enter the vehicle reachable space. Warning to the driver is issued in case of inminent danger. In addition, the vehicle's hood can be actively triggered when the collision is unavoidable, in order to mitigate its effect on pedestrians.





Lane Departure Warning System

LDWS provides accurate detection of road lane markings using a single microcamera. Distances to the left and right road markings are computed with centimeric accuracy. The system exhibits robust performance in all weather conditions, both at day and at night. Warnings are issued to the driver in case of inadvertent departure.