The aim of this work is to study how Deep Learning approaches solve safety problems in the interaction between autonomous vehicles and pedestrians. Three object detection algorithms based on Deep Learning are proposed. These different proposed algorithms can provide good results that can accurately predict accidents related to road safety. These algorithms are very close in terms of accuracy, precision and F1- score, which is why we chose the recall term in the comparison to focus on the false negative, which is the most important in our case. The algorithm chosen, which is ResNet50, shows the best performance in terms of recall with a value of 98.5%, i.e. an improvement of at least 1.1% compared with the algorithms included in the comparative simulations (CNN and VGG16). It can therefore used to be recognize and identify pedestrians. In conclusion, this study shows that Deep Learning approaches, in particular the use of the ResNet50 algorithm, offer a promising solution for improving safety in interactions between autonomous vehicles and pedestrians.
Collision avoidance in pedestrian-rich environments using deep learning
15.05.2024
1211869 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Pedestrian Collision Avoidance Using Deep Reinforcement Learning
Springer Verlag | 2022
|VEHICLE COLLISION AVOIDANCE SYSTEM WITH ENHANCED PEDESTRIAN AVOIDANCE
Europäisches Patentamt | 2016
|Vehicle collision avoidance system with enhanced pedestrian avoidance
Europäisches Patentamt | 2018
|