|
Print Friendly Format
Utilization of Leica ADS40 Digital Sensor data in the State of New South Wales (NSW), Australia: An Assessment of the Requirements of Software Technology and Methods for Rapid Response Mapping Applications
 Dr Dipak Paudyal ERDAS Queensland Australia dipak.paudyal@erdas.com David Abernethy Department of Lands Bathurst Australia david.abernethy@lands.nsw.gov.au Abstract: Department of Lands, New South Wales (NSW) has acquired a Leica ADS40 Digital Camera since 2007. The primary purpose of the sensor is producing statewide Orthophotos for a variety of Mapping and Environmental applications. The sensor is often required to provide assistance in Rapid Response Mapping applications such as in times of flooding, large rainfall and storm events, and to provide assistance during organization of large events. This paper will briefly examine the use of currently available feature extraction software and methods to provide for robust and reliable information extraction from ADS40 data during such events. This paper will focus on Rapid Response Mapping Application using one such automated/semi-automated software technology that is commercially available. This paper will also suggest future requirements for such technologies based on practical experiences gained during the deployment of the sensor for Rapid Response Mapping applications in NSW. 1. Introduction The use of imagery and remote sensing and photogrammetry techniques in mapping applications has increased manifolds in recent years. What is different now from what it was 10 years ago is that much of this data is also used in non-conventional mapping applications. By non-conventional we mean mapping applications that are not limited to production of standard topographic map sheets and contours etc. Mapping now has to be quick and accurate. The decision makers need the information as soon as possible. One such area where map-like information needs to be provided quickly is in area of disaster management or emergency management. We would like to call this as Rapid Response Mapping (RRM). RRM’s are typically required by governments and relief and rescue organizations for applications such as in storm damage assessment and management, flooding, bushfire events, volcano and earthquake and events such as oil-spill monitoring in ocean surfaces. We also notice that RRM efforts have capitalized on the advent of new sensor technology for imaging and newer software techniques in rapid production and dissemination of geospatial information. This paper discusses utilization of one such sensor, the second generation Airborne Digital Sensor called ADS40 and processing and delivery of data captured by this sensor to decision makers as a rapid response mapping application. 2. The ADS40 Sensor Leica Geosystems was first to introduce the Large Format Digital Camera to the market in 2001. Leica Geosystems has introduced two new sensor heads – SH51 and SH52 –for the ADS40 System. Users of this sensor are now able produce 5 Band co-registered imagery to cover applications in aerial surveying and airborne remote sensing. Time and costs are saved through simultaneous acquisition of high-resolution panchromatic, colour and colour infrared digital images. In fact color and panchromatic stereo-viewing is possible with SH51 images; and additionally, color-infrared stereo viewing with SH52 images. The images are captured as CCD lines each with 12000 Pixels. The size of the pixels is 6.5 Em X 6.5 Em. The field of view or swath angle is 64° with a focal length of 62.77mm. The sensor can simultaneously capture data in Forward Angle (27°), Backward Angle (16°) and at 2° Angle in nadir, with the viewing directions staggered by 3.25 Em. If required images can be captured up to 5cm ground sampling distance (GSD). The physical camera system and the sensor look angles that can potentially be used for image acquisition are depicted in Figure 1.  Figure 1: Second Generation ADS40 Sensor and Multispectral data acquisition system configuration. ADS40 is designed to achieve perfect co-registration of all multi-spectral bands by the utilization of Tetrachroid beamsplitter technology. ADS40 allows for the direct georeferencing of the acquired image data and accuracy of this (without the use of any additional control points) is often adequate for Rapid Response Mapping application, where it is the time that is often the critical factor for decision makers. In addition, ADS40 has very robust navigation systems on board that make it a very useful sensor in times of emergency use of the aircraft. An easy to use touch screen operation, highly automated flight control management system and integrated flight guidance systems are key components of this airborne system that make it very useful for variety of applications. 3. Feature Extraction from ADS40 data The NSW Department of Lands has officially documented the NSW Landscape using Aerial Photography since 1947. NSW Lands operate its own Cessna- 421c aircraft platform for data acquisition purposes. Many users in NSW are now aware of digital technology. The Department purchased Leica Geosystems manufactured ADS40 an Airborne Digital Sensor system in late May 2007. ADS40 system has provided NSW Lands department with an end-to-end mapping solution. The camera system captures and processes imagery at 50 cms Ground Sampling Distance (GSD) for statewide mapping programs, which is predominantly for producing orthophoto maps for the state of NSW. Multispectral imagery data are captured and utilized for variety of applications including Remote Sensing and GIS. The sensor is able to make an on-demand capture at greatly increased spatial resolution of upto 5cm through to a coarser resolution of 75cm if need be. This is what underlies the Rapid Response Capability of the sensor. This would mean that Rapid Response Mapping and dissemination of the information would be key use of this sensor statewide when required to be deployed in such a way. The following is a non-exhaustive list of events in which NSW Lands was requested to acquire Aerial Imagery primarily for RRM type applications as special projects. a. Asia Pacific Economic Co-operation Conference 2007, Sydney b. World Youth Day 2008, Sydney c. Western Sydney Storm Damage, 2007– State Emergency Services d. Northern NSW Floods event, 2008– coverage of towns affected by flood in 8-rural towns For the purpose of this paper we will be utilizing ADS40 data collected over areas in Western Sydney for Storm Damage and Northern NSW towns for the effectiveness of Feature Extraction of flooded areas for Rapid Response mapping applications. 4. A brief Review of Currently Available Feature Extraction Techniques The term “feature extraction” can encompass a broad range of methods and processes that allow user to extract features of interest from remotely sensed imagery. The underlying principle is that a remote sensing scene can be decomposed into several features that can be associated with remote sensing scene objects. Examples of such scene objects are vegetation types (forests and crops), urban features (such as roads and houses) and any other natural or man made objects that are present in the scene. An efficient feature extraction method is able to extract the features as “cleanly” and “accurately” as possible. There are potentially two broad methods of feature extraction. The first one of these is manual where the operator remains in control of the most of the initial and/or final stages of the extraction process. The second method involves automated and/or semi-automated process in which the operator provides some initial cues to the process and the operator continually tweaks or refines the process till an acceptable result is achieved. There are other groups of processes using photogrammetric process that could possibly be defined as “feature extraction” process but these processes are outside the scope of this paper. One of the reasons being the authors consider such processes as more appropriately manual “ feature collection” tools rather than “ feature extraction”. Given the scope of the work we will also only use the software tools developed by ERDAS to review the feature extraction techniques. The purpose of this study is to initially investigate and make a general assessment of currently available feature extraction techniques and make a general recommendation on the utilization and application of these techniques. Once this is established, the authors believe that similar software from other suppliers may be used to derive similar outcomes. Specifically, ADS40 data will be analysed for Rapid Response Mapping application using Imagine Objective, a semi-automated feature extraction software from ERDAS. 4.1 Methodology of Feature Extraction An object based feature extraction and classification software has been used to demonstrate the practical use and adequacy of this technology for Rapid Response Mapping Applications. Traditionally, the capture and update of such Mapping Applications are carried out by cost and labor intensive manual digitization process and surveying. But in Rapid Response Mapping applications time and reasonable accuracy that is required to the job at hand are considered critical. For images with coarser resolution the pixel is composed of several different objects thus making it amenable for treatment with traditional techniques such as supervised classification that are solely based on statistical methods that can handle mixed pixel scenario accurately. With the advent of higher resolution satellites traditional techniques of feature extraction and mapping using techniques such as supervised classification often do not provide accurate results. At higher resolutions pixel data can be treated more like a grouping of individual objects. As a result traditional methods fail to differentiate these objects purely on statistical basis. It is much easier to identify and map these features using object based approach rather than purely pixel based approach. The object of this paper is to evaluate Imagine Objective that is reported to blend the traditional image processing with computer vision and artificial intelligence through the use of pixel level and true object processing. The ultimate objective of this would be to emulate the human visual system of image interpretation. The algorithms are reported not only to perform raster contouring but also incorporate object and vector level processing to yield a spatially matched precise shape for each feature. This way it will be possible to produce GIS ready products such as smooth roads and squared-up buildings.  Figure 2: Second Generation ADS40 Sensor and Multispectral data acquisition system configuration. Operators are designed as plug-ins so that more can be easily added as required for specific feature extraction scenarios. Figure 2 shows the feature model building process as implemented in Imagine Objective. We can note the linearity of the workflow and that some nodes in the process can be skipped if so desired depending on the nature of the object to be extracted. The process starts with the identification of training and candidate pixels which are submitted to compute Pixel cue metrices to train the Pixel classifier. The output of the Pixel classifier is a Pixel Probability layer in which each pixel’s value represents the probability that is the feature of interest. An operator is then used to convert the Pixel Probability Layer into a Raster Object Layer. The Raster Object Layers are then subject to operations that perform mathematical morphology on the objects in the raster domain. The output is a new Raster Object Layer. The output from Raster Object to Raster Object Operator contains pixels that are grouped as raster objects which have associated probability metrices. Rater Object to Vector Object Operators is then used to convert objects from the raster domain to the vector domain. These operators take as input the Raster Object Layer from the previous steps and convert each raster object into a vector object, normally a polygon or polyline, and then produce a Vector Object Layer. The remainder of the process then deals with these vector objects. A class of operations performs operations on the vector objects and produces Vector Object Layer. These operators are often required to change the shape of vector objects by generalizing, smoothing etc Imagine Objective then extends the capability of more traditional feature extraction process by operating on Candidate Vector Objects. During the training phase vector objects that are representative of the feature of interest may be submitted to compute Object Cue Metrices to train the Object Classifier. During the automated extraction phase candidate objects from Vector Object Layer are submitted to the Object Classifier for query to measure how closely they resemble the training objects. The query result for an object will be recorded as an object attribute to be used in downstream processes. Cue Algorithms are then used to quantify human visual cues (such as shape, size orientation etc.) by computing cue metrices and the object level cues are computed on polygons and polylines. A class of operations is then performed on the Vector Object Layer output from the Object Classifier query and is intended for probability thresholding and for cleanup of the object polygon or polyline footprint. Output from Vector Object to Vector Object Operator contains the vector objects in a layer. This is the final result of the process. 5. Case Studies 5.1 Hale storm Damage Assessment and Mapping A total of 7 thunderstorms occurred in Sydney during December 2007, the highest number since 1959 when 11storms were recorded. The historic average is only 4 storms. Heavy downpours with storms early in the month caused local flash flooding in the inner west, Liverpool and northern beaches. A supercell thunderstorm on 9 December produced heavy rain, squally winds and giant hail up to 7cm size from Blacktown to Wahroonga with unconfirmed reports of 11cm size hail in Cherrybrook. In the wake of a severe hail storm on Sunday (9 December), all the utilities and other service providers were in standby. This included Gas, electricity, representative from insurance agencies. The purpose of the teams was to ensure that service was delivered in the most efficient way. Many houses had their roof damaged due to winds and hailstorm. These roofs were covered in Tarpaulin as soon as possible to minimize further damages. Urgent repair of the roof structures was of priority and NSW state government was keen to identify the properties that needed repair and assist property owners by monitoring the progress of such repairs by coordinating with the building insurers. This is where the NSW Lands Department that owns and operates ADS40 sensor was asked to provide assistance. The request of assistance came through State Emergency Services (SES) NSW. The primary reason of the assistance was to monitor the progress of the repair of the roofs by replacing temporary tarpaulin roof with fixed roof structures. In many cases this would involve the SES helping the residents to get on their repairs quickly by contacting insurance agencies to authorise the roof repairs urgently. Some members of the public and authorities were not comfortable on the speed of repair and wanted to monitor the situation so that the affected residents were assisted by the insurance companies in a timely manner. ADS40 was therefore flown on five different occasions so that progress of the roof repair could be monitored. The progress was tracked using manual methods by identifying the properties based on the presence and absence of the tarpaulin cover. The presence of Tarpaulin Cover at time t1 and absence at t2 would mean that the roof repair is complete and vice versa. The result was then overlaid over the property/parcel boundary to identify the owner and contact details. NSW state government authorities would then assist the owners by helping speed up the repair by getting into contact with insurance companies as deemed necessary. This paper has used the ADS40 data flown over Penrith (Western Sydney) on one such occasions (Dec 15th) to investigate the possibility of automating the feature extraction process to compare against the results obtained using manual methods that was used at the time. 5.2 Flood Damage Assessment and Mapping The State Emergency Services (SES) team put a request to NSW lands on Saturday, 5th of January 2008 in anticipation of major flooding event that was likely to occur in North East NSW following forecast from the Bureau of Meteorology. The main reason for flying the floods was for post event analysis. SES wanted to see where they may need more rescue boats, where hazardous materials may be exposed and for future planning. Also the imagery would prove to be valuable for clean up and checking which areas had become isolated. In some cases no ground control could be used because of the rush and unavailability of useable control. ADS40 Rapid georeferencing capability without having to use accurate control information came in very useful in this instance. In addition the sensor did an admirable job of flying through cloud and rain; not an ideal situation to fly the sensor. It has been brought to the attention of authors that SES and NSW Lands staff have had a significant praise for ADS40 for this. Imagine Objective was used to automatically extract flooded areas so that these can be subject to further GIS analysis as required. 6. Result and Discussions Imagine Objective was used to automatically extract tarpaulin covered roofs that needed urgent repair. Figure 3 shows the tarpaulin covered roofs mostly seen in blue, with some roofs covered with yellow tarpaulins. Figure 4 shows the automatically extracted roofs and that almost all the roofs are correctly extracted. The operator simply needed to overlay a shapefile of the property boundary over the imagery to identify and report the damaged roofs that needed urgent attention. In contrast to this Figure 5 shows the tarpaulin covered roofs that were missed using manual head-up digitizing extraction method (Omission Errors). Figure 6 shows errors of commission using the manual methods.  Figure 3: Roofs covered in Blue and Yellow Tarpaulins  Figure 4:Hail damaged roof extracted using Imagine Objective (in red)  Figure 5: Errors in Omission (Missed polygons are shown in yellow)  Figure 6: Errors in Commission (Incorrectly identified polygon are shown in yellow) Imagine Objective was used to extract the flooded areas as quickly as possible. A quick and reliable extraction is mandatory for a disaster management. Once the flooded areas are extracted then these areas can be subject to further GIS analysis as required. Figure 7 shows extraction of flooded areas during the flood event in Northern New South Wales in the Tweed Shire Council area. The flooded areas have been accurately identified.  Figure 7: Automatic Capture of flooded areas (Red Boundary) 7. Conclusions and Recommendation The hail storm data was set up on a particular web channel exclusively for the Western Sydney Hail Recovery Team for use in disaster management. As part of this research a possible improvement in the delivery mechanism was identified. A comprehensive solution featuring OGC/ISO standards compliant Service Oriented Architecture (SOA) for disseminating massive amounts of imagery and features using optmised Web Services is recommended. These web services then would automatically harvest and update data as these come in after processing and analysis. The provision of web services would also allow for rapid download and delivery of the data as required by end-users for Rapid Response Mapping Applications. It was also concluded the feature extraction technology currently available was generally adequate to provide reliable and accurate information for specific events for Hale Storm and flood management using ADS40 Imagery. 8. References:
|