[time]
2022-08-29T10:07:15+02:00

[track]
6

[team_name]
team708

[team_institution]
School of Software Engineering, Beijing Jiaotong University

[logolink]


[team_members]


[reference_person]
shuli zhu, xue yi, long zhang, liu feng, xueqi li, kejia li, jiayao liu

[reference_email]
zhushuli@bjtu.edu.cn

[description_short]
We split the Vehicle Dead Reckoning (VDR) problem into two sub-problems: one is the speed estimation, and the other is the heading estimation. As we all know the speed and heading of the vehicle per second, based on the initial position, we can calculate the position of the vehicle at the next moment. We use an LSTM-based time series model to infer vehicle speed. We design second-level mobile phone accelerometer and gyroscope features as model input to learn the speed change of the vehicle in that second. Of course, we can also set the learning target as absolute speed. We used a basic integral algorithm to estimate heading. Considering the jitter problem during the driving of the vehicle, we use the data of the accelerometer in the steady state to project the three-axis data of the gyroscope to the plane with the direction of gravity as the normal vector.

[description_long_link]
https://github.com/Juderer/IPIN22_Track6/tree/main/application

[publish_check_]
true

[results_check_]
true

[data_check_]
true

[pdf_check_]
false

[video_check_]
false