Video feeds#

Video feeds from different cameras:

frontal
vertical
stereo pair (left, right)
disparity map
depth map
depth from motion
occupancy grid

Frontal#

Frontal video feed parameters depend on recording and streaming configuration. The parameters for the default recording mode (4K UHD) are defined below.

Parameters	Physical/Simulated drone
Width (px)	3840/1280
Height (px)	2160/720
Format	NV21
FPS	30
Name	fcam_recording

Frontal video feed parameters for streaming.

Parameters	Physical/Simulated drone
Width (px)	1920/1280
Height (px)	1080/720
Format	NV21
FPS	30
Name	fcam_streaming

Vertical#

Video feed parameters from vertical camera.

Parameters	Physical/Simulated drone
Width (px)	320
Height (px)	240
Format	grey (8 bits)
FPS	60
Hfov (°)	53.71
Vfov (°)	40
Name	vcam_raw

Stereo pipeline#

The stereo camera allows to compute 3D environment information. (see here for more details). The next schema represents the different feeds available.

Stereo pair reference image#

Rectified video feed from the stereo right camera. The rectified image is computed using a pinhole camera model and stereo calibration.

Parameters	Physical/Simulated drone
Width (px)	1408/1280
Height (px)	720
Format	grey (8 bits)
FPS	30
Hfov (°)	110
Vfov (°)	72
Name	fstcam_right_rectified

Disparity map#

Each pixel of the disparity map represents the distance between the corresponding point in stereo left and right camera. The disparity is given in the rectified right image reference. This is computed by comparing visual difference between left and right picture. If the disparity cannot be computed, the values are set to 255 (not available).

Parameters	Physical/Simulated drone
Width (px)	176
Height (px)	90
Format	uint8
FPS	30
Hfov (°)	110
Vfov (°)	72
Range	[0 - 96]
Name	fstcam_stereo_disparity_filtered

Depth map#

Image containing depth information. The depth is computed from the disparity image. The depth is given in the rectified right image reference. If the depth cannot be computed, the values are set as NA (not available).

The depth information is the distance between the observed point and the image plane.

Parameters	Physical/Simulated drone
Width (px)	176
Height (px)	90
Format	raw32 (interpreted as float32)
FPS	30
Hfov (°)	110
Vfov (°)	72
Range	[30cm - 11m]
Name	fstcam_stereo_depth_filtered

Occupancy grid#

3D map of the environment as an evenly spaced voxel. Each containing the probability of presence of an obstacle.

Probabilities are stored as logodd :

logodd(p) = log( p /(1-p) )
Value between [-100- 100].
Value stored as int8.
Positive values mean that there is an obstacle in the voxel.
Negative values mean that the voxel is empty.
0 value means that the information is unknown.

The occupancy grid size is 64x64x32, the dimension voxel is 0.4mx0.4mx0.4m.

The occupancy grid translates in NED along with the drone, so that the drone position is always in the middle of the grid (32x32x16).

The occupancy grid orientation never changes with respect to NED frame.