University of Twente Student Theses

Login

Cyclist Maneuvers Prediction with Bicycle-mounted IMUs

Kaniscev, Ilja (2024) Cyclist Maneuvers Prediction with Bicycle-mounted IMUs.

[img] PDF
7MB
Abstract:Cycling is one of the most common types of transportation in the Netherlands, accounting for about a quarter of all types of transportation used bythe population in 2023. This type of transport continues to lead to deaths due to underdeveloped bicycle safety systems. Hence, the paper introduces new research to create an improved deep learning model regarding F1-score and advance notice time, predict cyclist maneuvers, and allow for a safe environment. Experiments were conducted using bicycle/cyclist-mounted IMUs and a smartphone device, which captured GPS data on 20 participants.This data is analyzed to identify potential indicative signs of pre-maneuvers. These signs are then utilized as contributing features in two prediction models: CNN-LSTM and CNN, expectantly predicting the maneuvers with substantial prior time and F1-score. Feature importance analysis is conducted to determine what features contribute most to prediction. Moreover, different combinations of prediction gaps and window sizes are tested on models to determine the optimal configurations. The final configured CNN model could predict the cycling maneuvers 1.2 seconds in advance and the window size of 1.6 seconds with an F1-score of 0.84. The final configured CNN-LSTM model achieved an F1-score of 0.83, predicting 1.2 seconds in advance with the window size of 2.6 seconds.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science, 55 traffic technology, transport technology
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/100838
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page