DACT: Dataset of Annotated Car Trajectories



DACT contains two subsets of annotated car trajectories data. The dataset contains 50 trajectories which cover about 13 hours of driving data. In DACT, we manually specified significant driving patterns by using an interactive framework. A significant driving pattern can be a turn, speed-up, slow-down, etc. An example of a segmented trajectory with meaningful driving patterns specified using ovals is shown in the following picture. The annotation process consists of a crowd-sourcing task followed by comprehensive aggregation phases. The aggregation is done by two different strategies: Strict and Easy. For the first one, we used some strict constraints to aggregate crowd-sourcing results, while we used flexible constraints to generate the second subset of DACT. More information about this dataset may be find here.


Please cite the following paper if you use this dataset:

Moosavi, Sobhan, Behrooz Omidvar-Tehrani, and Rajiv Ramnath. “Trajectory Annotation by Discovering Driving Patterns.” Proceedings of the 3rd ACM SIGSPATIAL Workshop on Smart Cities and Urban Analytics. ACM, 2017.

Format and Download

The data is presented in csv format, where we have two files, each based on one of the aggregation strategies. You can download these datasets from here. In each file, we have the following columns:

TripIdThe id of a trajectoryNo
TimeStepThe record number for a point of a trajectoryNo
TimeStampThe timestamp for a point of a trajectoryNo
SpeedThe ground velocityNo
AccelerationThe rate of change of speedNo
HeadingThe bearing which is a value between 0 and 359No
HeadingChangeThe change of bearing from the last observationNo
LatitudeThe latitude coordinate of GPS observationNo
LongitudeThe longitude coordinate of GPS observationNo
AnnotationExpert annotation which specifies the end point of a segmentYes
SegmentTypeThe type of a segmentYes

Application of Dataset

These datasets can be used to evaluate trajectory segmentation approaches, with the goal being to explore meaningful patterns, such as turn, speed-up, hard-brake, etc. Example of a research paper which used these datasets for evaluation of a segmentation algorithm can be find here.