The Airborne Science Initiative LiDAR Beach Survey; development and results

Transcription

1 International Global Navigation Satellite Systems Society IGNSS Symposium 2011, Sydney, NSW, Australia November 2011 The Airborne Science Initiative LiDAR Beach Survey; development and results P. J. Mumford (02) G. Nippard J. H. Middleton E. Kearney C. Cooke I. Turner M. Mole K. D. Splinter ABSTRACT The Airborne Science Initiative is a collaborative effort between the School of Aviation, School of Surveying, School of Civil Engineering and School of Biological, Earth and Environmental Sciences at the University of New South Wales. Harnessing the equipment and skills of the team, a series of beach surveys have been performed to gather data about beach dynamics over a period of six months. Long range weather forecasts have predicted that a major storm event will occur in 2011 with the development of a deep low pressure system in the Tasman Sea. Two beaches were chosen for the study; Narrabeen and Wamberal, in an effort to document beach erosion before, after and during a major storm event. This paper documents the design of the system, the data flow and the analyses techniques used. Results of initial accuracy testing are provided, along with other results available at the time of writing. Finally, future development and missions are discussed.. KEYWORDS: LiDAR, GPS/INS, beach, erosion, airborne mapping

2 1. INTRODUCTION Beach erosion affects many beaches in Australia. During storm events sand is often mobilised and moved offshore, resulting in loss of dry beach width and dunes and leaving upper beach structures (including buildings) vulnerable to direct impact of waves. To measure the change in a beach resulting from a storm event, surveys can be done before, during and after the event. Survey methods include measuring height lines along the beach with GPS equipped quad bikes and airborne survey using laser scanning equipment. Both methods were used in the study; however, this paper focuses on the airborne survey component. This paper is not intended to provide background to scanning laser survey methods and data processing, or to reference technical papers on this subject, rather to give a snapshot of the system and activities of the survey crew during recent missions. Airborne missions where flown from March to July to collect data over beaches from Manly in the South to Wamberal in the North. Flights where conducted on a monthly cycle to ensure pre-storm data was available. Additional flights where conducted during storm events whenever possible. Figure 1 shows the view from the aircraft looking South towards Palm Beach on one of the stormy missions. Data from Narrabeen flights on 14 th and 19 th of June and 13 th of July was used to test the accuracy of the system. 2. AIRCRAFT Figure 1. Storm over Palm Beach, Sydney. A Piper Seminole twin engine four seater, with a cruise speed of 290km/h and a range of 1600 km was used for the survey (Figure 2). The aircraft is owned and operated by the School of Aviation at the. It is used primarily as a twin engine trainer within the Department s program. An opening has been constructed in the left-hand luggage area as can be seen in Figure 3. The opening can be closed off for normal operation. The left-hand passenger seat is removed to make space for equipment. The conversion from trainer to survey aircraft can be performed in several hours, dependant on the actual setup. The opening is divided into three zones with baffles in between to reduce wind flow effects. From the front there is a laser scanner followed by an optical camera and finally an area used for flight guidance cameras and other equipment housed in an acrylic dome (Figure 4).

3 Figure 2. Piper Seminole VH-FRI in the UNSW Bankstown hanger. 3. SYSTEM DESIGN Figure 2. Sensor opening viewed from below. The sensor opening is designed for the flexible placement of equipment. The provision of optical domes and various mounting fixtures allows for different setups within the existing flight certification. The primary sensor used in the aircraft is a Riegl LMS240i laser scanner (also known as a Light Detection And Ranging LiDAR device). In addition there is a downfacing optical digital camera (Canon 5DMKII) and various small cameras used for flight guidance. Positioning and attitude determination data is provided by a Novatel SPAN CPT (Novatel SPAN n.d.). Real-Time-Kinematic GPS processing is used with a Radio Technical Commission for Maritime Services (RTCM) V3 differential correction data feed via a 3G modem from the CORSnet-NSW service provided by Land and Property Information (LPI 2011). The 3G modem is placed looking downwards in an optical dome together with a guidance camera. A GPS antenna is fitted on the upper airplane skin, directly above the SPAN device. The SPAN and LMS240i are axis aligned with a vertical offset.

4 Figure 4. View of LiDAR and camera fitted into opening. Figure 5. System setup viewed from luggage door. The LMS240i has a time-sync option that enables time-stamps derived from GPS to be output with the laser range and angle data. This allows synchronisation of position and attitude data (output by the SPAN) with the laser scanner data to generate geo-referenced point clouds. The 1 Pulse Per Second (1PPS) output from the SPAN is fed to the LMS240i, this is used to generate measurement time-stamps. The SPAN can generate time-mark files from an event trigger; this is used to time-synchronise camera images. The sensor equipment is mounted on a rigid platform. This platform is isolated from the aircraft frame with flexible gel mounts, designed to dampen the vibration caused by the engines. The platform and equipment setup is shown in Figure 5. Aircraft rated sealed lead-acid gel batteries provide power for the system. Both 12V and 24V are available. A sine wave inverter provides 240V power for the laptops. Laptop computers are used for configuration, control of and data logging from the sensors. Figure 6 gives a view from the rear luggage area and shows the large laptop computer connected to the LMS240i and SPAN devices. A smaller laptop placed in the front passenger area controls the cameras and is also used for flight guidance.

5 Figure 6. View of installation from the rear luggage area. 5. INITIALISATION AND OPERATION Initialisation steps for the LMS240i and SPAN CPT include: performing the LMS240i timesynchronisation; coarse alignment of the SPAN inertial sensors; connecting to the RTCM data feed from CORSnet-NSW; setting up the SPAN for RTK operation; and setting up SPAN logging options. Figure 7. LMS240i control application.

6 A custom application (depicted in Figure 7) is used to control the LMS240i setting and data logging. It is designed to be easy to use and to ensure correct operation with large buttons and visual status cues. This is required as operating complex tasks on a laptop during flight with the aircraft bouncing around and other distractions is difficult. 5. POST-PROCESS DATA FLOW The primary outputs of surveys are LAS files (LAS 2009). The binary file from the LMS240i is in the Riegl propriety.2dd format and needs to be converted to a format with known fields. Fortunately the Riegl application 3ddTOasc.exe can do this conversion. The application Kugeo takes the reformatted laser scan file and the position and attitude file from the SPAN and generates a point cloud file in either LAS or plain text (ASCII) format. Figure 8 shows screenshots of the file converter and Kugeo applications. LAS point cloud files can be opened with LAS file viewers such as LASEdit or imported into other spatial data tools such as ArcInfo (ArcInfo n.d.) or Envi (Envi 2011). The plain text point cloud files can be imported into data processing software such as Matlab for analysis and surface construction. Figure 8. Applications used for generating point cloud files from LMS240i and SPAN files. 6. ACCURACY TESTING Of primary concern is the point cloud horizontal and vertical position accuracy. In an effort to improve on previous accuracy testing, a target was constructed in the shape of a V with retro-reflective film on a black plastic sheet. This 2 X 2 m target was placed on Narrabeen beach (pointing North) and surveyed with RTK GPS prior to the flight. The V target can be found in the processed data cloud and coordinates extracted. There is, however, a pointing error associated with this simple technique that limits the accuracy test to about 0.5 m in the horizontal. Figure 9 depicts the target in the point cloud and in a photo in Figure 10. The highlighted point with attached coordinates is the best choice for the apex of the V, but clearly a laser point will rarely fall right on the apex. Point precision available from the LASEdit query tool is provided at cm level only. At typical mission heights and speed, laser points are approximately 1 m apart. Fortunately the vertical accuracy estimate is better as the

7 target on the beach could be considered nearly flat over the span of two laser points, reducing the pointing error. Figure 9. Point cloud with V target. Figure 10. V target on Narrabeen beach. The V target was deployed on Narrabeen Beach on 3 occasions, 14 th and 19 th of June and 13 th of July. Table 1 shows results for these days. The second column is the latitude offset between the V apex point in the LiDAR cloud and the surveyed point expressed in metres. The third column is longitude offset and the forth is ellipsoidal height offset. Result statistics are shown in Table 2. The second row is the average offsets for latitude, longitude and height. The third row is the standard deviations of the offsets. The vertical component of GPS measurements is generally considered to be worse than the horizontal component due to the satellite geometry. However, the horizontal component of a cloud point has errors associated with aircraft attitude estimation errors, and the SPAN IMU LMS240i alignment and offset estimation. Analysis of data in the horizontal and vertical components helps separate these effects on accuracy.

8 Date / file / height Delta lat in m Delta lon in m Delta H in m 14-6 narra_2 250m narra_3 240m narra_4 200m narra_1 190m narra_2 190m narra_1 260m narra_2 250m narra_3 250m narra_4 310m narra_5 310m narra_6 320m narra_6 460m narra_7 460m narra_9 480m Table 1. Differences between surveyed V and cloud apex point. Delta lat in m Delta lon in m Delta H in m Average Std Table 2. Average and standard deviations of deltas. On the beachfront at Narrabeen is two unit blocks with flat roofs. One LiDAR point on the rooftop is selected from the point cloud of each survey. The height of the roof is compared on 9 surveys over 4 dates from 16/06/2011 to 26/07/2011. This provides an indication of the repeatability of the survey method in the vertical component. Table 3 shows the results. The measured heights span 43cm with a standard deviation of 15cm. file Height of roof (m, Ellipsoidal) narra_2_ narra_3_ narra_4_ narra_2_ narra_3_ narra_2_ narra_1_ narra_3_ narra_5_ average Standard dev 0.15 Table 3. Height of Narrabeen unit rooftop over 9 surveys. In summary, while horizontal accuracy can only be claimed at around 1 m due in part to the pointing error, vertical accuracy is better than 0.5 m in all cases, and much better (decimetre level) in about half the cases (Table 1).

9 7. FUTURE A new sensor is currently being built; a state-of-the-art hyper-spectral camera to complement the existing sensors. This will broaden the range of scientific missions that can be flown. Further accuracy testing and calibration of the system will be conducted. One proposed method involves the extraction of planes (from building surfaces) from the LiDAR point cloud with known (surveyed) geometry. These can be used to calibrate the SPAN IMU to LMS240i rotational alignment. Another proposal is to construct a larger V target that will allow the reconstruction of a V of best fit in the LiDAR point cloud to better estimate the position of the apex and therefore reduce the pointing error. The Kugeo processing software will be further developed with an improved interpolation algorithm. This is to help produce smoother and more accurate point clouds, particularly during turbulent flights where there can be significant dynamics within the 0.5 second position and attitude output rate from the SPAN. It is hoped that the system will be further used to advance environmental science research in NSW. REFERENCES NovAtel SPAN, NovAtel, accessed 12 August 2011, Land & Property Information (previously The Land and Property Management Authority LPMA) 2010, LPI, NSW Australia, accessed 12 August 2011, LAS, 2009, American Society for Photogrammetry and Remote Sensing REF (ASPRS), accessed 12 August 2011, ArcInfo, Esri, accessed 12 August 2011,, Envi, 2011, ITT Visual Information Solutions, accessed 12 August 2011,