Detection of the Intention of Cyclists
The dataset consists of video recordings of bicycles and small electric vehicles performing different maneuvers. The recordings contain different types of bicycles and different riders. Additionally the dataset contains recordings with an e-scooter and an e-board.
Different maneuvers such as starting, stopping, and turning with hand signal left and hand signal right were considered. These are shown from different distances and perspectives. The only exception is the e-board for which the dataset consists mainly of regular riding due to the difficulty of balancing for inexperienced riders. The dataset shows different movement patterns of the respective road users and can be used in research in the field of intention detection and motion understanding.
The data set consists of 75 recordings and 75 csv files. The recordings contain drive-by motions of one or more persons performing the respective movement at different positions. Each recording is accompanied by a csv file which for each frame of the recording contains information about the type of road user (bicycle, e-scooter or e-board) for the currently executed movement as well as the corner points of a bounding box for each road user. Some examples of the dataset are shown in the following figure.
A detailed description of the data available in the csv is attached to the dataset in the form of a readme.
The dataset consisting of the videos, csv files and related information (e.g. readme) is available under
The dataset was created as part of the project
This project was funded by the German Federal Ministry of Transport and Digital Infrastructure from resoruces of the the National Cycling Plan 2020 under grant VB1905.