University of Twente Student Theses
Cycling Maneuver Prediction Using Bicycle-mounted Motion Sensors
Smit, Gijs de (2023) Cycling Maneuver Prediction Using Bicycle-mounted Motion Sensors.
|
PDF
1MB |
| Abstract: | There has been a growing number of road accident fatalities involving cyclists. Timely prediction of a cyclist's maneuver can be useful to proactively avoid potential accidents. This research focuses on utilizing bicycle-mounted motion sensors for maneuver prediction. The goal is to find out to what extent the maneuvers of left turn, right turn and cruise can be predicted using a Convolutional Long Short-Term Memory Neural Network (CNN-LSTM). A literature survey and preliminary experiment are conducted to motivate sensor placement on the handlebar, which is compared with placement on the frame. An experimental setup is designed where 20 participants performed the three cycling maneuvers on an artificial intersection outlined with cones. The data collection method consists of a custom Android app to record IMU data of a smartphone and a Thingy:52 sensor. The model input is optimized by removing unnecessary features, adding processed features and low-pass filtering. The final model was able to predict the cycling maneuvers 0.25 s in advance using a 1 second time window with an F1-score of 0.92. Lastly, future research can be done into combining various sensors and placements to increase the prediction time gap and improve classification performance. |
| Item Type: | Essay (Bachelor) |
| Faculty: | TNW: Science and Technology |
| Subject: | 54 computer science |
| Programme: | Advanced Technology BSc (50002) |
| Link to this item: | https://purl.utwente.nl/essays/96458 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page