Abstract Quickbird Vs Aerial photos in identifying man-made objects

Transcription

1 Abstract Quickbird Vs Aerial s in identifying man-made objects Abdullah Mah abdullah.mah@aramco.com Remote Sensing Group, emap Division Integrated Solutions Services Department (ISSD) Saudi Aramco, Dhahran Copyright Saudi Aramco 2005 Quickbird satellite imagery with 61 cm spatial resolution (pansharpened multispectral data) can be used to substitute aerial s of 1:20,000 scale and larger but still inferior in quality than aerial s of 1:10,000 scale and smaller. Man-made objects such as earth mounds are identifiable on Quickbird imagery but much clearer on aerial s in smaller scales (< 1:10,000). Wire line fences with only 2-3 mm thick wires cannot be seen on both the Quickbird and aerial s but can be identified indirectly from their shadows. The shadows of wire line fences are much clearer on aerial s of 1:10,000 scale and smaller than on Quickbird satellite imagery. Very small scale aerial s (1:5,000 or smaller) can be used to easily identify earth mounds and wire line fences but not necessarily the most cost effective. The study compared Quickbird satellite imagery and aerial s of scales 1:30,000, 1:20,000, 1:150,000, 1:10,000 and 1:5,000 for the cost effective and optimum resolution images in identifying man-made features such as earth mounds and wire line fences and other man-made objects for early encroachment signs.

2 (Keywords: spatial resolution, pan-sharpened, encroachment) Quickbird Vs Aerial s in identifying man-made objects Introduction Since the launch of the first Landsat in 1972, resolution of satellite images have improved immensely. From 79 m spatial resolution of the multispectral data of Landsat MSS and 30 m of Landsat Thematic Mapper, satellite imagery spatial resolution has been gradually improved to four m for IKONOS and in 2001, to 2.44 m for Quickbird. Using the pan-sharpening technology, spatial resolution of multispectral data are today enhanced to the resolution of their respective panchromatic. Hence, multispectral data of IKONOS are often spatially enhanced from four m to one m and in the case of Quickbird from 2.5 m to 0.61 m using their counterpart panchromatic data. Quickbird with pan-sharpened 0.61 m resolution images covering a swath width of 16.5 km is preferred by many over the conventional aerial s of one m resolution for land application and GIS use. There are other factors that have to be considered in choosing a particular type of data for specific projects. The paper discusses the pros and cons of using Quickbird opposed to aerial s of five scales for an encroachment project where object identification is crucial. 2

3 Objective The objective of the study is to find out whether Quickbird high resolution satellite imagery can substitute aerial s in the identification of man-made objects for an encroachment project. Test site The test site is located in the Middle East (Fig. 1). The area coverage varies based on the scales of the aerial s. Aerial s of scale 1:5,000 cover the least area with a width of one km by 4.5 km length positioned diagonally along the NE direction. Aerial s of scale 1:30,000 cover the largest area (six km x 17 km). Fig. 1. Comparison of area sizes of aerial s with five different scales. The area is located in the Middle East. 3

4 Data Quickbird with 61 cm resolution panchromatic and 2.44 m multispectral data and aerial s of 1:30,000, 1:20,000, 1:15,000, 1:10,000 and 1: 5,000 are used in the study. Quickbird Quickbird collects a swath width of 16.5 km with a scene covering 16.5 km x 16.5 km. At 61 cm resolution; buildings, roads, bridges and other detailed infrastructure become. The imagery can be used for a wide range of applications, from assessment and management of land, infrastructure, to natural resources. The present study used Quickbird imagery of an area captured in April, Table 1: Characteristics of the Quickbird satellite sensor Launch Date October 18, 2001 Launch Vehicle Boeing Delta II Launch Location Vandenberg Air Force Base, California, USA Orbit Altitude 450 km Orbit Inclination 97.2º, sun-synchronous Speed 7.1 km/second - 25,560 km/hour Equator Crossing Time 10:30 a.m. (descending node) Orbit Time 93.5 minutes Revisit Time days depending on Latitude (30º off-nadir) Swath Width 16.5 km x 16.5 km at nadir Metric Accuracy 23 m horizontal (CE90%) 4

5 Digitization Resolution Image Bands 11 bits Pan: 61 cm (nadir) to 72 cm (25º off-nadir) MS: 2.44 m (nadir) to 2.88 m (25º off-nadir) Pan: nm Blue: nm Green: nm Red: nm Near IR nm Aerial s The aerial s used in the study were captured on March 3, 2005, with a Leica (152 mm lens) camera mounted on a fixed wing plane. From different heights of the plane, the aerial s were taken in the morning with 1:30,000 scale s taken at about 9:00 a.m. Smaller scale aerial s were subsequently taken and 1:5,000 scale s were taken last, about an hour later. The negatives of the s were then scanned using Leica DSW600 scanner of 2032 dpi resolution. Table 2: Area coverage, number of s for aerial s of each scale Data Scale Area No of File size Acquired type coverage s date Aerial 1:30,000 ~ ~ 950MB 3 Mar, 05 sqkm Aerial 1:20,000 ~ 45 sqkm 6 ~ 950MB 3 Mar, 05 Aerial 1:15,000 ~ 25 sqkm 6 ~ 950MB 3 Mar, 05 Aerial 1:10,000 ~ 15 sqkm 7 ~ 950MB 3 Mar, 05 Aerial 1:5,000 ~ 5 sqkm 11 ~ 950MB 3 Mar, 05 Quickbird 61cm res ~ sub ~ 330MB 4 April, 05 sqkm scene 5

6 Methodology To find out whether Quickbird can substitute aerial s in the identification of man-made objects for encroachment projects, Quickbird and aerial s of the study area are co-registered to the same datum and projection for comparison. Buildings, agricultural farms, earth mounds and wire fences are identified on the Quickbird and on the aerial s of five scales and compared to see which type of image is the best to identify the above mentioned man-made objects. In choosing the most suitable type of image, cost effectiveness, technical complication and processing time are also considered. Data preparation The blue, green and red bands of the multispectral data (2.44 m resolution) of Quickbird imagery were pan-sharpened (0.61 m resolution) using BROVEY transformation [((target XS band)/(xs1+xs2+xs3))*panchromatic] and a Natural Color look RGB321 was constructed. The raw aerial s were registered against Pan-sharpened Quickbird (61 cm) of the area using the Cubic third order polynomial rectification and nearest neighbor re-sampling method. Numerous ground control points (> 10 GCPs) were used in each rectification with RMS less than one. 6

7 All the Quickbird and aerial s are projected in WGS84 and NUTM39. Discussion Scale and image/ground resolution comparison of aerial s When the aerial s are registered against the Quickbird, they were automatically adjusted to the image/ground resolutions (Table 3). When visually compared between features (buildings, high voltage power lines posts, earth mounds, etc) aerial s of 1:20,000 scale have the closest image/ground resolution to the Quickbird Pan-sharpened 61cm resolution image, and are slightly better. The 1:20,000 aerial s are estimated to have an image/ground resolution of ~ 50 cm. The discrepancy between the computed and observed image/ground resolutions may be due to the fact that the aerial s were most likely scanned using a higher resolution scanner than was necessary. For image comparison purposes the computed image scales are maintained but for interpretation purposes the estimated image/ground resolutions are used. Table 3: Computed and estimated resolutions of the aerial s referenced to Quickbird Data type Scale Automaticall y adjusted image/groun d resolution Estimated image/ground resolution (visual Comment 7

8 Quickbird 61 cm Aerial Aerial Aerial Aerial Aerial from rectification against Quickbird comparison) Pan-sharpened 1:30, cm ~75 cm Estimated against 1:20,000 aerial 1:20, cm ~50 cm Closest to Quickbird but slightly better 1:15, cm ~37.5 cm Estimated against 1:20,000 aerial 1:10, cm ~25 cm Estimated against 1:20,000 aerial 1:5,000 6 cm ~12.5 cm Estimated against 1:20,000 aerial Availability, cost and spectral comparison between Quickbird and aerial s Quickbird is taken from the BGIS 2000 sensor on board the satellite with a fixed sun-synchronous orbit that has a one to 3.5- day repeat cycle (depending on latitude) whereas aerial s, being airborne, are more flexible and can be taken on any day but subject to the logistics of the data vendor. Both have to depend on the weather. Images taken on a heavily clouded or rainy day will be 8

9 of no use. Quickbird price can be obtained from the Digital Globe data vendor. Costs of aerial s vary depending on the scale and area coverage, but they are generally more expensive than Quickbird for the same area coverage. The cost to process aerial s is also more expensive as the area coverage per aerial is generally much smaller than the area coverage of a Quickbird scene (16.5 km x 16.5 km) and hence will require many aerial s to cover a study area. Spectrally, Quickbird has more spectral dimensionality as it has four multispectral bands covering range plus near infrared of the electromagnetic wavelength and one panchromatic whereas conventional aerial s cover only the range with RGB 3 bands. Furthermore, since aerial s are taken at a much lower altitude (a few km above the ground against 450 km satellite altitude of Quickbird), geometric and radiometric distortions are much higher than the Quickbird images. Before deciding which type of image to use in a specific project one needs to look into the advantages and disadvantages of each type of data to achieve the objectives of the project. Comparison of detectable features on Quickbird and aerial s Buildings with flat cement roofs are on Quickbird and aerial s of all scales with smaller scale s (1: 10,000, 1:5,000) showing clearer edges of the buildings (Fig. 2). 9

10 Fig. 2. Building, trees comparison on Quickbird and aerial s Trees with large crowns (> three m diameter) are recognizable on Quickbird and aerial s of all scales with smaller scale s showing much clearer definition of the trees (Fig. 2). As for smaller trees (< one m crown diameter) and bushes, they appear only as dark circular features but only on smaller scale aerial s. High voltage cable posts are not on Quickbird nor on large scale aerial s. Positive recognition of objects on images and aerial graphs requires a ground resolution of three to five times smaller than the object size (Hall 1997). Only the shadows of the posts are on the Quickbird and larger scale aerial s as the length of three pixels on the image (ground resolution) is larger than the width of the metal rods that are used 10

11 to construct the high voltage cable posts. Metal rods used to construct the cable posts are about 50 cm width for main frame and less (6-12 cm width) for supporting rods (field measurement - high voltage cable posts in a suburb). As smaller scale aerial s have widths of about 75 cm to 37.5 cm for three pixels (25cm x 3 = 75cm for 1:10,000 scale aerial and 12.5cm x 3 = 37.5 cm for 1:5,000 scale aerial ) the widths of three pixels for the two finer small scale aerial s are close to or smaller than the width of the main frame metal rods. Hence high voltage cable posts range from barely to, on those two small scale aerial s (Fig. 3). Fig. 3. High voltage power lines, car tires comparison on Quickbird and aerial s 11

12 Car tires appear as dark circular features on aerial s of all scales where on smaller scale aerial s one can see the circular shape of the tires clearly (Fig.3). Earth mounds piled up by bulldozers are on Quickbird and aerial s of all scales and are much clearer on smaller scale aerial s (Fig. 4). Fig. 4. Earth mounds comparison on Quickbird and aerial s Wire fences are not on Quickbird nor on any of the aerial s but only their shadows are barely on Quickbird and larger scale aerial s, and on smaller scale aerial s (Fig. 5). This is because the wire fences are a few mm in thickness and the fence posts are less than 10 cm in width. These dimensions are much smaller than the three pixel width of Quickbird and all different scale s. 12

13 Fig. 5. Wire fences comparison on Quickbird and aerial s The visibility of the features on Quickbird and aerial s of all scales are summarized in the table below. Table 3: Visibility of features on Quickbird and aerial s Feature Quickbird Aerial 1:30,000 Building s Trees Earth mounds Car tires High Voltage Cable posts Visible Dark tone but recognizable Visible but blur Very dark shades of trees Aerial 1:20,000 Visible Dark tone but recognizable Aerial 1:15,000 Visible with clear edges Dark tone but recognizable Visible Visible Visible Clearly Appear as black Appear as Appear as Appear as dots black dots black dots black dots posts but not the posts posts but not the posts posts but not the posts posts but not the posts Aerial 1:10,000 Visible with clearer edges Dark tone but Clearly Visible posts and posts Aerial 1:5,000 Visible with very clear edges Very clearly Very clearly Clearly Both shadow of the posts and posts 13

14 Wire fence wire fence barely wire fence barely wire fence barely wire fence barely barely wire fence are wire fence Conclusions and recommendations Identifying features such as permanent buildings, large scale agriculture farm using Quickbird and aerial s will be straight forward even on the 1:30,000 scale aerial s. To identify features that may indicate early encroachment signs such as earth mounds piled up by bulldozers and car tires spreading out regularly as boundaries to claim land, and wire fences erected for the same purpose will be challenging. Among the three possible early encroachment signs, earth mounds and car tires are easier to identify but the most challenging identification task will be to identify wire fences. Earth mounds and car tires can be identified on Quickbird and aerial s of scales 1:20,000 and smaller. On aerial s of scale 1:30,000 earth mounds and car tires are not as clear as on the Quickbird and aerial s of 1:20,000 scale and smaller. As for the wire fences they can be identified only indirectly by their shadows. Aerial s of 1:5,000 scale are the best when compared to with the aerial s of larger scales to indirectly identify wire fences. Due to cost and increased time consumption in processing smaller scale aerial s (present study example: more time to process 11 aerial s of scale 1:5,000 covering smaller area (1 x 4.5 km) than to process 7 aerial s of 1:10,000 scale covering (2 x 6 km)), and the fact that 14

15 aerial s of 1: 10,000 can still be used to reliably identify early signs of encroachment features, aerial s of scale 1:10,000 are recommended to be used for the detailed study of areas that are prone to encroachment. The study concludes that for broad feature encroachment such as large buildings and agricultural farm, Quickbird can be used and substitute aerial s. For early encroachment signs such as to identify earth mounds, car tires and wire fences, it is better to use smaller scales aerial s such as 1:10,000 or smaller. One has to bear in mind that using aerial s of smaller scale like 1:5,000, though better in resolution will require more computer resources and processing time as the number of aerial s will increase significantly, hence increasing the overall cost. References Hall, R.C Post war strategic reconnaissance and the genesis of project Corona. In R.A. McDonald (ed.), Corona: Between the Sun & the Earth, The first NRO reconnaissance eye in space, p American Society Photogrammetry and Remote Sensing. Quickbird satellite imagery characteristics: