Naslov
Neural Network-based Prediction of Pedestrian Crossing Behavior at Uncontrolled Crosswalks

Autori
Topić, Jakov ; Škugor, Branimir ; Deur, Joško ; Ivanović, Vladimir ; Tseng, H.Eric

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Skup
5th International Conference on Smart Systems and Technologies (SST)

Mjesto i datum
Osijek, Hrvatska, 19.-21.10.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
vehicle-pedestrian interaction, autonomous vehicle, neural network, pedestrian crossing behavior, prediction

Sažetak
This paper deals with prediction models of pedestrian crossing decisions meant to be used within autonomous vehicle safe speed control strategy. The emphasis is on stochastic models capable of capturing the inherent uncertainty and variability typically present in real pedestrian behavior. Instead of predicting whole pedestrian crossing trajectory, for the purpose of simplicity only ego-vehicle relevant quantities are targeted for prediction, i.e., a pedestrian entry time to and exit time from the crossing area, as they determine the pedestrian time occupancy of the conflicting crossing area shared by both agents. To this end, two independent feedforward neural network models are employed, aimed to predict conditional probability distributions of the aforementioned quantities in dependence on the current vehicle- and pedestrian-related states. Finally, the proposed models are parameterized and verified for a single vehicle/single pedestrian case, based on data generated from numerous simulations of available game theory-based pedestrian model, where the vehicle is driven in an open-loop manner.

Izvorni jezik
Engleski

Znanstvena područja
Strojarstvo

POVEZANOST RADA