UAV LiDAR
Reasons to UAV LiDAR Easy mobilization Low operation costs Through mass production of sensors & GNNS available to manufacture even in small companies Data quality compared to ‘professional’ systems not as high but sufficient to many targets Lot of small projects and new applications like supplementary mapping, project monitoring etc. Compared to photogrammetric point clouds can see ground also on covered areas Special targets like power lines Lot of potential new users
Mapping categories in SWOT Coverage Small Field UAV Aerial Remote Sensing LiDAR Photogrammetry High Flexibility Low Satellite Large
Limits of UAV LiDAR Reliability : ’all UAVs crush in one day’ Operation time 15…30 minutes => small areas, short corridors Permissions: < 150 m altitude < 500 m from the pilot Visible during operation Depends on national laws/ rules
System Concept Needed software: Survey planning Navigation GPS antenna SPAN integrates GNSS with Inertial Measurement Unit (IMU) => Exact position and orientation of system continuously Needed software: Survey planning Navigation Post-processing Data calibration and classification Image processing Board + memory for navigation and data saving Scanner with 32 channels Camera
LiDAR sensors Velodyne announced yesterday that the company is set to begin high-volume manufacturing of sensors. Enabled by strong sales of the VLP-16 “Puck” LiDAR sensors, the company has invested in a production-line style manufacturing process for existing and forthcoming products, including a new automotive sensor that is currently in the preliminary design concept phase. Lower Costs for All Velodyne Sensors First, Juchmann explained that Velodyne is scaling up their production to allow the company to meet the price demands of the automotive industry. The automotive market requires sensors that cost even less than the VLP-16, he said. “The big driver to get the cost further down, that’s the automotive industry. Obviously, if you buy a car and you have to spend $8,000 on an additional sensor, that just doesn’t work. The car industry is looking for below $1,000, even well below $500 for a sensor.”
http://www.phoenix-aerial.com/information/lidar-comparison/
Principle of orientation/ location IMU + GNSS Defines the location and orientation of the system XYZ: WGS84 geocentric Camera Distance RTK GNSS Receiver PostProcessing
Data Accuracy Error Sources: Other Incorrect trajectory solution (poor GNNS) No or poor data calibration Final matching to control points missing Insufficient data classification Other Co-ordinate transformation Matching to poor Geoid surface
Position Improvement by Post-processing Flight line = GPS + IMU (original) = GPS + IMU + RTK = post-processing = GNSS position = rejected GNSS position
Data Calibration/ TerraMatch real theoretical Heading Flight direction Roll Pitch Solution: To compare overlapping point clouds to find correction parameters Surface comparison Tie line comparison Before After
Target Co-ordinate System Planar XY In general bases on Gauss-Kruger etc. Mathematical form between east/north co-ordinates and longitude/ latitude Elevation Bases on geoid Zero elevation medium sea water level No mathematical form
Nature of Laser Point & Accuracy Laser pulse expands with increasing distance: Divergense 3 mrad => footprint 15 cm from 50 m altitude (Velodyne VLP-16) Laser points are not points In quality control one compares the distance from ground surface (created from laser point) to control points. Not the accuracy of single laser point.
Phoenix Aerial ALS32-LiDAR Data Processing by Terrasolid software Goals: To test the calibration and data quality of Phoenix Aerial ALS32 system by Terrasolid software Operator:Phoenix Aerial Systems Date: 4.11.2015 System: Phoenix Aerial ALS32 with Sony Alpha A6000 camera + Velodyne 32 scanner Altitude: 30-35m Data set: - Trajectories, 200 Hz - Total 139 904 361 points - Pictures tota 249
Data Processing Overview Software TerraScan – for reading-in datasets in to geographical blocks TerraMatch- Data calibration TerraPhoto – Image processing In calibration only a limited area used Continuous trajectory was cut and the pieces were numbered All returning curves were removed Some hard surfaces and buildings on the ares
Calibration and quality improvements Hard surface classification => removes all except the medium level point from total Measure Match to get a value of the difference of hard surfaces from different trajectories Noise before data calibration Coloring by 32 LiDAR channels of the HDL-32E scanner. Coloring by trajectories Noise before calibration: Trajectory Points Magnitude Dz 11 1095751 0.0856 -0.0726 12 843659 0.0819 +0.0943
Calibration by ‘tielines’ To correct heading, pitch, roll and dz of the system TerraMatch fit lines along points from overlapping trajectories. Compares the differences to find correction parameters Applies the corrections to the whole data set Point coloring after corrections Coloring by 32 LiDAR channels of the HDL-32E scanner. Coloring by trajectories Noise after calibration: Trajectory Points Magnitude Dz 11 1096744 0.0415 +0.0225 12 842999 0.0372 -0.0293
dz fluctuation corrections Last step Define the dz fluctuation by time Apply the corrections to the whole data set Point coloring after corrections Coloring by 32 LiDAR channels of the HDL-32E scanner. Coloring by trajectories Noise after correction Trajectory Points Magnitude Dz 11 1097056 0.0368 +0.0124 12 843071 0.0328 -0.0161
Smoothen the points An iterative process, which uses local small surface area to lift and lower the points to improve the surface geometry Point coloring after corrections Coloring by 32 LiDAR channels of the HDL-32E scanner. Coloring by trajectories Noise after smoothening: Trajectory Points Magnitude Dz 11 1097045 0.0268 +0.0109 12 843073 0.0229 -0.0141
System Analyses Scanning pattern is wide and point density high. Suits well for mobile mapping where distances are short. Due to the scanning pattern each target produces several echoes, of which a part is perpetual, a part oblique => a part of the total noise. Oblique hits cause a part of the noise ; limit max 30 m distance. Fan-like Scanning pattern in airborne LiDAR Scan angle left and right Scan angle along trajectory
Conclusions In flight planning sufficient overlap to get enough perpetual hits for calibration. Scan angel < +/-20 degrees Hard surfaces and building roofs needed for calibration With careful calibration one could reach in maximum 2 cm accuracy