RPAS Photogrammetric Mapping Workflow and Accuracy

Transcription

1 RPAS Photogrammetric Mapping Workflow and Accuracy Dr Yincai Zhou & Dr Craig Roberts Surveying and Geospatial Engineering School of Civil and Environmental Engineering, UNSW Background RPAS category and operation regulations RPAS photogrammetric mapping workflow Factors affecting product accuracy Summary 1

2 Fixed Wing vs. MultiRotor Advantages Simple structure High speed Long duration Disadvantages High mechanical complexity Low speed Short flight time Disadvantages Large take-off & landing space Not able to hover Inflexibility to carry different sensors Advantages Vertical take-off and landing Flexibility with different sensors Able to hover and stare 2

3 Weight classes: micro: gross weight < 100 g very small: gross weight kg small: gross weight 2-25 kg medium: gross weight of kg large: gross weight >=150 kg RPA selection: Job nature mapping, 3D modelling or inspection Takeoff weight light weight, less risks Sensor s weight or size Investment budget 3

4 RPA Standard Operating Conditions (SOC): VLOS (visual line-of-sight) in daytime only Under 400ft (120m) AGL (above ground level) 30m away from people Not in prohibited or restricted area Not within 3NM (5.5km) of a controlled aerodrome Not over populated areas Only fly one RPA at a time *Details in Advisory Circular AC Operation category: Excluded RPA No ReOC or RePL required Included RPA ReOC & RePL required 4 OC Operator s Certificate, PL Pilot s Licence

5 Design flowchart to determine eligibility as an excluded RPA 5

6 6

7 ReOC Application 1. A company with ABN/ACN 2. An employee to obtain RePL 3. Nominate a chief pilot 4. Prepare an Ops manual (CASA template) 5. Apply for ReOC 6. Chief pilot skill assessment by CASA 7. ReOC granted 8. Chief pilot is responsible for all operations 8

8 RPAS Optical Sensors RGB compact cameras RGB DSLR cameras NIR cameras Multispectral cameras Red edge 9

9 Image Distortion and Camera Selection Distortion caused by lens: Consumer-grade cameras large distortion Geometric-cameras small distortion and geometrically stable Distortion caused by moving camera: Electronic (rolling) shutter large distortion Focal plan (mechanical) shutter small distortion Leaf (mechanical) shutter no distortion (blurry?). Camera selection: Avoid rolling shutter Leaf shutter right choice Focal plan shutter for low speed flight 10

10 Project Planning Location (urban, rural or remote?) Site terrain, vegetation (rich texture on images?) Area of survey Google Earth KML Preferred time frame weather permits? Survey in standard operation conditions? Obtain permission from CASA required? Geometric accuracy requirement? GSD (ground sampling distance), terrain, flying height AGL? Site accessibility? Launch and landing area? Land owner consent GCPs (ground control point) required and survey accuracy? Flight planning. 12

11 Flight Planning (1) All photogrammetric measurements are based on overlapped images in order to obtain 3-dimensional object geometry 13

12 Flight Planning (2) Check Can I Fly CASA.gov.au Apply for permission if the flights are not in SOC Weather permit (raining / windy)? Flight height / image resolution / GSD setup Image overlap lateral and longitudinal Time of the day to fly (image quality) Set drone safety features (eg return home, ceiling, geofence) Trained observer on site RPAS firmware and apps updated Pre-flight checklist ready GCP survey pre- or post- flight Post-flight checklist ready 14

13 Flight Planning (3) 15

14 Image Quality Control 1. RPAS ground speed (motion blur) 2. Flight height ( GSD and image coverage) 3. Oblique angle ( GSD and image coverage) 4. RPAS vibration (gimbal) 5. Camera and lens quality (image distortion) 6. Camera shutter speed (motion blur) 7. Camera sensor resolution (GSD) 8. Camera aperture (depth of field out of focus) 9. Light condition (sunny shadow, cloudy under exposure) 10. Surface reflection (coal stockpile under exposure) 11. Large elevation change (GSD variation affect matching) 12. Time of day to capture images (e.g. high wall shadow) 17

15 GSD variation due to oblique angle Same sensor size Different ground coverage Different GSD Vertical vs Oblique images 18

16 GSD variation due to terrain changes (1) 19

17 GSD variation due to terrain changes (2) Constant flight H above take-off location (120m) Different camera H above ground ΔH=80m (40m 120m) Large GSD variation: 2cm at hill top, 6cm at bottom No oblique images 20

18 large image overlaps improve product accuracy Parallel flight paths (normal overlaps 70%L & 70%) 21 Crossover flight paths (= very large overlaps)

19 Image format no effects on accuracy JPEG format images Raw (CR2) images NIR images RGB images 22

20 Windy condition affects accuracy Flight tracks off designed paths due to strong wind. Less matched feature points in some areas 23

21 Water surface Water surface cannot be mapped or precisely surveyed photogrammetrically 24

22 Multiple flights planning - By planning software - By mission areas - No gaps or with small overlaps 25

23 Camera shutter type effects Rolling shutter (Phantom3) Electronic 1 st curtain (Sony QX1) 26 Leaf shutter (Canon compact)

24 Point cloud RTK check string 500m Point cloud validation (1) 400m RTK (30 sec/epoch) check strings on hard surface areas 27

25 Point cloud validation (2) Single flight + Nadir images Flight H = 120 m GCPs = 6 Image overlaps = 80% Number of Images = 110 Point cloud RTK string points: o Mean = -6.0 cm o Standard deviation = ±5.7 cm 28

26 Point cloud validation (3) crossover flights + Nadir images Flight H = 80 m & 120 m GCPs = 6 Image overlaps = 80% Number of Images = 220 Point cloud RTK string points: o Mean = -5.6 cm o Standard deviation = ±5.4 cm 29

27 Point cloud validation (4) 2 flights + oblique images Flight H = 120 m GCPs = 6 Image overlaps = 80% Number of Images = 330 Point cloud RTK string points: o Mean = +2.0 cm o Standard deviation = ±4.8 cm 30

28 Summary The following aspects need to be considered for the best practice of conducting a RPAS aerial mapping project. Accuracy depends on GSD Optimal σ xy = ±1 GSD; σ z = ±1.5 GSD 70% Image overlaps 90% Number of GCPs 5 GCP RTK 30 epochs (best with bipod) Oblique Images improve accuracy significantly Time of the day = light cloudy or mid-day (less shadow) Large elevation variation = oblique images + variable flight H Use a gimbal to stabilise camera (reduce vibration blur) Slow ground speed (reduce motion blur) High resolution optical sensor (small GSD) Leaf shutter lens (avoid rolling shutter effects) High quality camera (less sensor or lens distortion) 31