The large collections of datasets for researchers working on the Simultaneous Localization and Mapping problem are mostly collected from sensors such as wheel encoders and laser range finders mounted on ground robots. The recent growing interest in doing visual pose estimation with cameras mounted on micro-aerial vehicles however has made these datasets less useful. Here, we provide datasets collected from a sensor suite mounted on the “pelican” quadrotor platform in an indoor environment. Our sensor suite includes a forward looking camera, a downward looking camera, an inertial measurement unit and a Vicon system for groundtruth. We propose the use our datasets as benchmarking tools for future works on visual pose estimation for micro-aerial vehicles.
Five synchronized datasets - 1LoopDown, 2LoopsDown, 3LoopsDown, hoveringDown and randomFront are created. These datasets are collected from the quadrotor flying 1, 2 and 3 loop sequences, hovering within a space of approximately 1m x 1m x 1m, and flying randomly within the sight of the Vicon system. These datasets consist of images from the camera, accelerations, attitude rates, absolute angles and absolute headings from the IMU, and groundtruth from the Vicon system. Images from the first 4 datasets are from the downward looking camera and images from the last dataset are from the forward looking camera. Synchronized datasets for the calibrations of the forward and downward cameras are also provided.
More details of the acquisition of this dataset are given in our paper:
- Quadrotor flew 1 closed loop trajectory (downward looking camera) 1LoopDown
- Quadrotor flew 2 closed loops trajectory (downward looking camera) 2LoopsDown
- Quadrotor flew 3 closed loops trajectory (downward looking camera) 3LoopsDown
- Quadrotor hovering within a space of approximately 1m x 1m x 1m (downward looking camera) hoveringDown
- Quadrotor flying randomly within the line-of-sight of the Vicon system (front looking camera) randomFront
- Calibration images (front looking camera) calibrationFront
- Calibration images (downward looking camera) calibrationDown
- Transformation matrix that relates downward looking camera to quadrotorT_Quad_Cam (Down)
- Transformation matrix that relates front looking camera to quadrotorT_Quad_Cam (Front)
- Downward looking camera intrinsic values intrinsic (Down)
- Front looking camera intrinsic values intrinsic (Front)
- ARToolKitPlus Markers configuration file ARTK Config
- Matlab file for benchmarking pose estimation algorithm with Vicon groundtruth (Cosine Similarity) benchmark.m
Note that the quadrotor frame (body frame) refers to the coordinate frame that we measure the Vicon readings. There is a separate IMU frame where the accelerations readings from the accelerometer are measured with respect to. The IMU frame is -4cm in the z-axis of the downward looking camera and it is taken to have the same orientation as the body frame. There is also an Inertial frame which is the fixed world frame. See the following figures for illustrations of these coordinate frames.