PRACTICAL ACCURACY TEST OF DIGITAL CAMERA EMERGE DSS Kikuo Tachibana Tadashi Sasagawa GIS Institute, PASCO Corporation 1-1-2, Higashiyama, Meguro-ku, Tokyo, Japan [email protected] [email protected] Mohamed M.R. Mostafa Applanix Corporation 85 Leek Cr. Richmond Hill, Ontario, Canada L4B 3B3 [email protected] Yuzo Tanahashi Leica Geosystems K.K 2-3-3, Uchikanda, Chiyoda-ku Tokyo, Japan [email protected]

ABSTRACT Since 2002, PASCO Corporation has been using two POS AV direct georeferencing systems together with Leica RC30 aerial film cameras. Several evaluation projects have been carried out, which resulted in proving that aerotriangulation is no longer essential and that direct georeferencing systems allow orthophoto production and medium/small scale topographic mapping (e.g. map scale of 1/2,500) without aerotriangulation or ground control. On the other hand, for large scale mapping (e.g., map scale of 1:1,000 and 1:500), due to GPS error sources, insufficient accuracy is obtainable when using direct georeferencing only. However, it has been proven that sufficient accuracy is achievable using POS-assisted block adjustment with minimum number of ground control. Nowadays, several commercial digital systems are available among which is the Digital Sensor System (DSS). The DSS is a fully integrated digital system that consists of a compact 4kx4k medium format digital camera with an embedded POS AV and a flight management system. The DSS is capable of capturing color and color infrared imagery, with a GSD of 0.05 m to 1.0 m. Since the DSS is compact, light, and easy to install, it can be used in several applications such as disaster mapping, archeological mapping and many other emergency and fast response projects. To evaluate the practical accuracy of the DSS, a test flight has been flown in Japan by PASCO Corporation in February, 2003. Onboard a helicopter, the DSS was used to acquire a 0.05 m GSD images (photo scale 1/5,500). In this paper, the performance of the DSS using the concept of POS-assisted bundle adjustment for large-scale mapping is presented in some detail.

INTRODUCTION Evaluation of the accuracy The Digital Sensor System (DSS) is a fully integrated airborne mapping system, which includes a direct georeferencing system, a medium format digital camera, a flight management system, and an azimuth mount. Developed by Applanix. Leica Geosystems is a non-exclusive reseller of the DSS. To evaluate the system accuracy for large scale mapping, A 5cm GSD images were captured by the DSS mounted on a helicopter in Japan. The 5 cm GSD resulted in an image scale of 1: 5555. Accuracy evaluation of the direct georeferencing approach is conducted and reported in this paper. The accuracy and characteristics of the direct georeferencing approach have been well studied using POS AV on a Leica R30 cameras for two years onboard a fixed wing aircraft. However, in the project at hand, the DSS has been mounted onboard a helicopter which has different dynamics than that of the fixed wing aircraft. This will be highlighted in the subsequent sections of the paper Photogrammetric accuracy evaluation has also been done by comparing the coordinates of checkpoints observed in the image models using different approaches. ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

HARDWARE CONFIGURATION Table 1 shows the specifications of the DSS camera component. Note that in the project at hand, a 55mm lens was adopted to capture color images. The DSS images are georeferenced by an embedded POS AV system which specification is shown in Table 2. Figure 1 shows the Helicopter deployed in this project which is AS350B (Aero Spatial). Table 1. Specifications of DSS DSS Spec. Array size Pixel size Filters Lens（Made by ZEISS）

4092 x 4077 Pixel (36.8 x 36.7 mm) 9 μm Color or color IR 35mm (FOV 55 deg) 55mm (FOV 37 deg)

Shutter speed Shutter cycle time

1/125～1/4000 sec 4 sec or 2.0 sec

GSD Image smear

0.05～1.0 m No up to 200 Knot (360km/h)

Table 2. Specifications of Direct Georeferencing System Specification Position (m) Velocity (m/sec) Roll & pitch (deg) Heading

Accuracy of post processed data 0.05～0.3 0.005 0.008 0.015

Figure 1. DSS installed camera box (Left) and Helicopter AS350B (Right) In general GPS antenna should be installed at the best position to avoid signal obstacle, but in the present case the antenna was placed at upper side of the window of the cockpit from inside temporarily since modification of the Helicopter just for this specific experiment study was not possible. Figure 2 shows GPS antenna installation in the cockpit. ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Figure 2. Temporary GPS antenna installation

The Test Site The test site is located at Toyonaka City of the Osaka Prefecture. Accurate and well distributed GCPs are available throughout the city. Almost all GCPs lie on manhole covers. The cover of manholes is about 40cm by 40cm square shape and the height difference from GCP to cover is also measured. This cover can be identified and observed easily in the image model. Highly accurate GCP coordinates were provided by Toyonaka City Office.

Flight configuration The configuration of the test flight is shown in table 3. Table 3. Configuration of the test flight TEST SITE Toyonaka City Reference Station Yao Airport Static Initialization Focal Length 55mm (H/B=3.7) Resolution (GSD) 0.05m Req. Min. Speed 73km/h (60% Overlap assumed) Flying altitude 306m Image scale 1/5,556 Captured area 204m Base interval 81m 60% lap Strip interval 142m 30% lap No. of strip 6 strips＊ No. of image 25*6=150 images＊ Block size 900m x 2,000m＊ ＊ 40 images of 5 strips (about 700m x 1,000m) are used in the photogrammetric evaluation.

EVALUATION OF DIRECGEOREFERENCING SYSTEM INSTALLED AT THE HELICOPTER Acquired data Figure 3 shows the trajectory of the test flight. A triangle at the lower right corner of Figure 3 is shown to depict the location of the GPS reference station, which is located at The Yao Airport. The upper left side of Figure 3 shows the Toyonaka test area. The average distance between the GPSreference station and the test area is about 24km. Figure 4 shows the variation of roll during the flight, where the Y-axis represents the roll angle in degrees while the X-axis represents the GPS measured in seconds of week. Figure 4 confirms that the roll angle experienced by the helicopter is much less when compared to that of the fixed-wing aircraft. ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Figure 3. Trajectory of helicopter and the reference station

Figure 4. Variation of ROLL value during the flight Figure 5 illustrates the condition of the captured GPS signal on the helicopter. The GPS data was collected at 2 Hz. The Y-axis shows GPS satellite number and X-axis shows GPS time. The color shows the elevation angle. The Tics on each line are the occurrence of cycle slip. In case of fixed wing aircraft, the banking angles at the turns are usually kept at about 15 - 20 degrees. This sometimes causes cycle slips to occur on the lower satellites but it is possible to have continuous signal acquisition for higher elevation satellites. On the other hand, in this flight, there were so many cycle slips even for the higher elevation satellites. The main reason was that the antenna was placed inside the cockpit window as shown in Figure 2 In summary, the situation for data capturing was not optimum due to this GPS antenna installation issue and cycle slips could have been avoided if the antenna is installed at the appropriate position even if it is installed under the rotor.

ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Figure 5. Condition of captured GPS signal

Data Processing Results and Analysis Post processing of direct georeferencing data was done using the POSPac software. This software package performs data extraction, Kinematic GPS analysis, and best estimation of trajectory and estimation of exterior orientation parameters. The raw GPS/IMU data were extracted to specified format and GPS kinematic analysis was carried out. Figure 6 shows variation of DOP where Y-axis is DOP, X-axis is GPS time. Although DOP is around 3 throughout the flight period of time, there are some degraded epochs caused by cycle slip.

Figure 6. PDOP During the Flight During Kinematic GPS analysis, various parameters such as applied satellites, elevation mask and error estimation for phase data should be set properly based on the conditions of the data. As shown in figure 5, there are several cycle slips and the solution was degraded when lower satellites were used during data processing. Thus, the elevation mask was set to 23 deg. Figure 7 shows the residuals of forward/reverse (time related) of Kinematic GPS solution. Although it is not absolute rule to estimate the final accuracy using those residuals, but it can be used as a quality control measure to check ambiguity resolution issues or blunders. Due to small number of satellites at the start of reverse analysis the residuals between forward and reverse show large difference. In this case, the forward analysis did not show any problem. The final antenna position was obtained by computing the weighted mean value of the forward and reverse solutions.

ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Figure 7. Residuals between Forward/Reverse solutions Figure 8 shows the plot of quality factor as one of indicators of quality control. Y-axis is the quality factor (1:Fixed solution, 2: stable float solution, 3: iterated float solution, 4: Less than DGPS). There are float solution parts but the relatively stable solution was obtained throughout the flight mission.

Figure 8. Quality factor of solution Figure 9 shows the standard deviation of positions derived from Kinematic GPS solution. It is around 5cm for the entire part and a stable solution is achieved. Best estimation of trajectory process is done using the Kinematic GPS solution and the IMU data. This process was done by POSProc software module and the final position and attitude parameters were estimated by the smoothing process using POSProc. Inconsistency of GPS and IMU solutions is also checked at this stage and the final adjustment was followed. Any problems which could not be detected at Kinematic GPS analysis sometimes can be detected at this stage of the data processing using POSProc analytical tools. Although the degradation of GPS solution due to the rather frequent cycle slips because of the poor antenna installation was a concerned, the GPS data processing was completed without any problem.

ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Figure 9. Plot of position standard deviation Figure 10 to Figure 12 show the RMS values of the final solution. As shown in Figure 10 to Figure 12, the final solution accuracy in the order of a few centimeters. It seems that the influence of cycle slips has been properly handled by the processing software.

Figure 10. RMS value of final solution (X)

Figure 11. RMS value of final solution (Y)

ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Figure 12. RMS value of final solution (Z)

PHOTOGRAMMETRIC ACCURACY EVALUATION Photogrammetric Data Results and Analysis To evaluate the photogrammetric positioning accuracy of the DSS, The GCPs of Toyonaka City were observed on the stereo models and compared with the DSS-derived coordinates. Forty images from 5 strips were used and the block adjustment was performed. The system used for measurement was Leica Geosystems’s SocetSET. The bundle adjustment was done using Applanix POSCal.. Although POSCAL is provided to determine boresight misalignment of the coordinate system between IMU and imaging sensor, those parameter are treated as unknown same as simultaneous block adjustment of Direct georeferencing data and photogrammetry measurement data. Camera Self Calibration and datum shift also can be treated as unknown parameters. Provided Radial distortion data was used as it is. Focal length and principal point values were also used with a standard deviation of 10 microns.

Figure 13. Photo stations and GCP/CHECK point A total 16 of GCPs were observed. Figure 13 shows 40 photo stations and GCP/CHECK points. Table 4 shows the results of the block adjustment. In table 4, (1) all points used as check points (since photogrammetric observations were also used, it is different as compared to the result of Direct Georeferencing only); (2) only one point at the center (310541) is fixed as GCP, (3) 4 corner points (2056, 30114, 30142 and 30283) are fixed as GCP and the rest as check points.

ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Table 4. RMS and maximum residuals of CHECK points

dX (m) (1) Without GCP

RMS

(check points16)

Max

(2) 1 point fixed

RMS

(check points15)

Max

(3) 4 points fixed

RMS

(check points12)

Max

dY (m) 0.04 0.11 0.04 0.10 0.06 -0.09

dXY (m) 0.07 -0.14 0.07 -0.14 0.05 -0.09

dZ (m) 0.08 0.15 0.08 0.14 0.07 0.10

0.15 -0.36 0.13 0.30 0.17 -0.38

Evaluation of the results In figure 14, in case of (1), all the 16 points were used as CHECK point and the RMS values are 8cm for planimetric and 15cm for height, respectively. Although one point shows a rather lager residual, it is reasonably good. In the case of (2) and (3), 1 GCP and 4 GCPs, the RMS values are also 8cm, 13cm and 7cm, and 17cm, respectively.

Figure 14. Check point residuals (1) Without GCP (Upper Left), (2) 1 GCP (Upper Right), (3) 4 GCPs (Lower Left)

EVALUATING THE DSS AGAINEST THE MAPPING ACCURACY STANDARDS IN JAPAN: For the mapping at the scale of 1/500, the official mapping standard in Japan requires an accuracy of 0.25m standard deviation for both planimetric and height values. The results achieved in the current project fulfilled this specification. For this block, the datum transformation using GCP is not needed and almost same accuracy was achieved even it has minimum GCPs. It seems that the EO parameters derived from the direct georeferencing system had high and homogenuous accuracy. In the simultaneous bundle adjustment, it was used as high quality observations which resulted in making the block geometry more stable and stronger. Figure 15 shows radial distortion and image shift caused by distortion and offset of principal point.

ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession

Distrtion(micro

R addialD istortion 0 -50 -100 -150 -200 -250 -300 -350

2

4

6

8

10 12 14 16 18 20 22 24 26

R adius(m m ) R addialD istortion

Figure 15. Radial distortion and the effect of these parameters

CONCLUSIONS AND FURTHER ISSUES The present case study confirms that the Digital Sensor System (DSS) is capable of delivering directly georeferenced digitally-acquired images for large scale map production. Even though the total number of images acquired by the DSS is relatively higher than that acquired by a film camera, this will not pose any problem because of the rather automated data post-processing capabilities currently available. Additionally, it has been confirmed that POS AV direct georeferencing system delivered accurate exterior orientation parameters albeit the existence of frequent cycle slips because of the temporarily poor antenna installation onboard the used helicopter It is recommended to study the effect of lens changes on the stability of the interior orientation. Additionally, the gyro stabilized mount controlled by the direct georeferencing system is needed to maintain enough over and course lap.

REFERENCES K.Tachibana et al., June 22 2001, JSPRS (GPS/IMU Direct Geo Reference Systems Test) Sasagawa et al., July 3 2002, APA [Direct Geo Reference （GPS/IMU status and issue） Sasagawa et al., Nov 14 2002, JSPRS [Direct Geo Reference Automatic Aerial Triangulation for large scale mapping] Sasagawa et al., Nov. 4 2003, JSPRS [Test of Direct Geo Reference with GNET ] Tachibana, K., Sasagawa, T. and Okagawa, M. 2003. Practical Accuracy Test of Direct Georeferencing Using 30 Sec. GPS Stations. Proceedings of the ASPRS Annual Conference, Anchorage, AK, May 5-9.

ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession