WPI VSLAM Dataset
Data Download
- This dataset was collected on the second floor of the Atwater Kent Lab, WPI, Worcester, MA, USA.
- For the project details please visit Our Lab Github Repo
- The collected dataset in Rosbag format. [Download: 49.7 GB]
- The sensor extrinsic calibration file between OS1-64 Lidar and Intel Realsense T265 camera. [Download: 67.7 MB]
- The camera intrinsic calibration file for Intel Realsense T265 camera. [Download: 26.5 MB]
- The camera-imu calibration Rosbag with undistored images for Intel Realsense T265 camera. [Download: 443MB]
- The 3-hours static IMU Rosbag for Intel Realsense T265 camera. [Download: 811MB]
Example Demos on MATLAB:
- Performant and Deployable Stereo Visual SLAM with Fisheye Images
- Estimate Camera-to-IMU Transformation Using Extrinsic Calibration
WPI Traffic Light Dataset
Training Data
- The training data are collected in Worcester, MA, USA during summer.
- ROI Samples: Extracted from frames in Frames_GT_wHolder. Can be used for training classifier. [Download: 77.8 MB]
Training Data [Optional]
- Training Data [Optional]
- These data are optional and may be helpful. E.g., if one wants to investigate not only the traffic lights, but also the surrounding areas.
- Many, if not all, the frames in Frames_GT_wHolder and Frames_GT_Lights are the same.
- Frames_GT_wHolder: Whole frames with label. Each folder is a type of traffic lights with their holders, and only that type is labelled (Ground truth contain the traffic light holders). A frame contains multiple types of traffic lights may appear in different folders. [Download: 10.5 GB]
- Frames_GT_Lights: Whole frames with label. These sequences are labelled for individual traffic lights of all types. Sequences 8,9,10 do not have traffic lights. [Download: 1.52 GB]
Test Data
- The test data are collected in Worcester, MA, USA during winter.
- Test: Whole frames with label. These sequences are labelled for individual traffic lights of all types. Sequences 18,19,20,21,22,23 do not have traffic lights. [Download: 3.06 GB]
Our Demo (compared with HOG features without tracking)
WPI Pedestrian Dataset
Data and Label
- The training data are collected in Worcester, MA, USA.
- In our paper, 19 sequences (seq3,6,9,...57) are used for test, the rest 29 sequences are used for training.
- Data and Label. [Download: 4.56GB]
Our Demo (Using CCF)
WPI Lane Keeping Dataset
Data
- The data are collected on part of I-84, I-90, I-290.
- The camera parameters, images, speed and steering wheel angles are provided.
- Data. [Download: 15.8GB]