ASSESSMENT OF SMARTPHONE CAMERA IN CLOSE-RANGE DIGITAL PHOTOGRAMMETRY Hoyong Ahn 1 Chuluong Choi 1 Yeon Yeu 2* 1 Department of Spatial Information Engineering, Pukyong National University, 45 Yongso-ro, Nam-gu, Busan, South Korea, hyahn85@naver.com 2 Spatial Information Institute, Pukyong National University, 45, Yongso-ro, Nam-gu, Busan, Korea *Corresponding author: singstars@naver.com ABSTRACT This study is intended to analyze accuracy of smartphone image in determining 3-dimensional location for approximated objects before photo survey system using smartphone is developed, and then evaluate its usability. First of all, unbalance type distortion model is used to determine lens distortion coefficient for verifying camera. Although lens distortion patterns of Galaxy Camera and Galaxy Note2 indicate different trends, bundle adjustment RMS deviation less than 1 mm is shown for all cases. Results depending on software used for image processing showed standard deviation within ±5 mm for all cases, which is believed to have excellent result for determining 3-dimensional locations. Lastly, based on result to calculate statistical value for residual of each method with observation value of check point performance by assigning performance of check points from Total station as optimized value, relatively higher deviation than X and Z direction, to the direction of Y that is a distance of photographing, occurred. As seen from above explanations, it's expected that smartphone is enough to be used to determine 3-dimensional locations in terms of accuracy. KEY WORDS: Smartphone, Accuracy, distortion model, Camera Calibration INTRODUCTION Recently, as low priced high resolution CCD (charge-coupled device) and CMOS (complementary metal-oxide -semiconductor) sensors are developed, demand of digital camera continue to rise. Because of this, technologies of digital cameras have been widened. In addition, thanks to easy acquisition of digital camera image, technologies of digital camera are widely used from computer vision to very specialized area such as numerical photogrammetry (Akca, D., 2009) Especially, use of high resolution and non-measurement camera continues to rise for numerical photogrammetry (Fraser, C.S., 1997) However, as most of commercial cameras are not designed for purposes of numerical photogrammetry, most critical issue for measurement purposes in camera is to correct distortion contained in camera lens (Habib, A., et.al, 2003) In order to correct this distortion, camera calibration is necessary and accuracy of observation for photogrammetry is directly related to quality of sensors and correct modeling for internal facial expression. Smartphone collectively indicates mobile device combining functions of mobile phone such as personal information management which was previously provided by PDA and it is also equipped with camera, GPS, accelerometer and other sensors for measurement such as magnetic measurement sensor. Recently as smartphone is widely used, researches on built-in sensors from smartphone are actively progressing. This study is carried out to photograph approximated object and acquire its 3- dimensional location information using a camera supporting high resolution image. METHODS Two kinds of cameras used for are Samsung GC(Galaxy Camera) and Samsung GN2(Galaxy note 2) Samsung GC consists of 1.4 GHz Quad-core CPU, 1GB internal RAM, 8 GB internal memory (extended Micro SD memory up to 64GB). Samsung GC consists of 16-Mpixel (4608*3456) and F.L has a range of 4.1 to 86.1 mm. GC is running by the operating system Android 4.1.2 (so called by Jelly Bean)
GN2 consists of 1.6 GHz Quad-core CPU, 8GB internal RAM (extended Micro SD memory up to 64GB) and GN2 produces images by 8-M (3264*2448) pixel while it is running under Android 4.1.2 just like GC. (Fig. 1.) Figure 1. Smart Camera used in this study Galaxy Note2 (Right), Galaxy Camera(Left) Distinctive optical features for camera need to be addressed in order to carry out approximated photogrammetry using non-measurement purpose camera. This process is called camera calibration. Data that is acquired by camera calibration include focal length, locations of major points and lens distortion factors. Calibration sheet is photographed from No 4. Location as shown in Fig. 2. In this case, a total of 8 sheets are photographed by 90 rotating camera clockwise from 4 directions and rotating it 90 counter clockwise at No 3 and No. 4 location. Next, this study analyzed accuracy of location for each case using smartphone which is completely calibrated. For this purpose, this study observed coordinates of GCP using Total station and GPS at No 14 location of rooftop of a building in Pukyong National University. CX-105 of SOKKIA is used for Total station and Topcon GNSS receiver GRX 1-U is used for GPS. 6 points among GCP are used for base points for conversion into targeted coordinates and 8 points are used as check points to evaluate accuracy. (Fig 3.) Figure 2. View of Calibration Sheet Figure 3. View of Accuracy Test Site RESULTS AND DISCUSSION Camera Calibration Calibration of camera & IO orientation are calculated from perspective geometric model using bundle adjustment (Brown, 1971) As results of Bundle adjustment, Table 1 shows result of Camera Calibration. F.L.(focal length) is calculated to be 4.486mm for GC and 3.666mm for GN. P.P(principal point) coordinate showed P.P.x & P.P.y is (+)0.048mm, (+)0.041mm for GC while it showed P.P.x & P.P.y is (+)0.003mm, (-)0.009 mm for GN2.
Table 1. The results of Camera Calibration GC GN2 Camera type EK-KC120S SHE-ES250S Sensor type BSI-CMOS CMOS Sensor size 1/2.3" 1/2.3" CMOS Width, Height (mm) 6.451, 4.838 4.569, 3.427 Image Width, Height(pixel) 4608, 3456 3264, 2448 Image Width, Height(pixel) 4608, 3456 3264, 2448 Output format JPEG JPEG F.L. (mm) 4.486 3.666 P.P.x(mm) 0.048 0.003 P.P.y(mm) 0.041-0.009 Pixel size (X,Y)(μm) 1.4 1.4 Image coordinate RMSE (pixel) 0.493 0.515 Adjusted target coordinate RMSE (mm) 0.146 0.258 K1 4.21E-04-0.00897 lens Radial Distortion Lens Tangential Distortion K2-3.92E-06 0.00112 K3 0 0 P1 5.66E-04 0.000382 P2-5.32E-04 0.000121 RMSE for GC's image coordinate is 0.493pixel and that of GN 2 is 0.515pixels. This indicates lens distortion by perspective projection is small judging from the fact that RMSE value for lens of Multi lens construction GC is small. Fig. 3 shows radial distortion curve based on camera calibration for each project. This applied radial distortion that is determined by photographer and the amount of radial distortion is within 25 μm for all cases. Figure 4. Lens distortion curve for Galaxy Camera(Right), Galaxy Note2(Left)
Accuracy assessment of smartphone Camera This study calculated RMSE value of Image Coordinate using bundle adjustments of ERDAS LPS 9.2 and GCP in 14 base points that are measured by Total Station. In addition, for remained GCPs, this study calculated geometric strength of stereo image using RMSE of Image Coordinate calculated by check points. RMSE of The Image Coordinates calculated by triangulation is shown to be a little higher than 0.5pixel. In case of inverse triangulation, RMSE to the Y-direction is significantly improved regardless of calibration method. This is a standard collimation direction of camera. Like this, this study analyzed 3-dimensional locations for 8 check points according to each method assuming that distortion coefficient of camera is internal facial factor and then analyzed its accuracy by comparing those locations with coordinates determined by Total Station. Table 2. The Difference of GCP positioning between Total Station and image-triangulation (Unit :m) GC GN Point Vx Vy Vz Vx Vy Vz 1 0.012 0.060 0.103-0.049 0.110 0.042 2-0.004 0.012 0.004-0.020 0.151 0.102 3-0.107 0.034-0.122-0.017 0.019-0.010 4-0.028 0.005-0.070-0.015-0.017-0.040 5-0.027 0.032-0.037-0.060-0.033-0.112 6-0.047 0.021-0.056-0.040 0.012-0.077 7-0.024 0.002-0.046 0.007 0.062 0.040 8-0.062-0.023-0.109-0.056 0.058-0.118 9-0.032-0.016-0.068-0.004 0.012-0.005 10 0.020-0.030 0.041-0.050-0.013-0.077 11 0.035-0.037 0.085-0.004-0.102 0.075 12 0.041-0.130 0.129-0.061-0.084-0.041 13-0.059-0.130-0.089 0.015-0.074 0.085 14 0.015-0.009 0.021-0.021-0.053-0.001 15-0.093-0.066-0.046-0.011-0.020-0.002 Min -0.107-0.130-0.122-0.061-0.102-0.118 Max 0.041 0.060 0.129 0.015 0.151 0.102 Mean 0.040 0.040 0.068 0.029 0.055 0.055 Std.Dev 0.029 0.041 0.037 0.021 0.043 0.040 Assuming that performance of check points by Total Station in Table. 2 is the most probable value and performance of check points determined by each project, statistical calculation for residual for each method is shown in Table 2. Maximum values for standard deviation are ±0. 041 m, ±0.060 m and ±0.129 m for each direction while minimum values are ±0.061 m, ±0.102 m and ±0.015 m for each direction and average values are ±0.029 m, ±0.041 m and ±0.037 m. In summary, relatively higher residual occurred to the Y-direction rather than X and Z-direction. CONCLUSIONS This study is conducted to evaluate usability of approximated photogrammetry for camera installed in smartphone that is widely used recently. This study used GC and GN for research purposes. It calibrated camera of smartphone prior to evaluation of accuracy of images in smartphone.
After camera calibration, bundle adjustment RMS deviation is within 1 mm for all cases. As the result of statistical calculation for residuals for each method assuming that performance of check points by Total Station is the most probable value and performance of check points determined by each project is observed value, RMS deviation is shown to be within 0.2 m for each X, Y and Z- direction. This is not just result equivalent to that of measuring camera or results of studies on nonmeasurement cameras or even better. As summarized above, although utilization of smartphone camera is expected to be enough in terms of accuracy to determine 3-dimensional locations, efforts to reduce deviation would be necessary. And it's also expected that this study provides excellent outcomes from studies in connection with UAVs, applications and Wi-Fi in the future. ACKNOWLEDGEMENT This work was researched by the supporting project to educate GIS experts. This research was financially supported by the Ministry of Education (MOE) and National Research Foundation of Korea (NRF) through the Human Resource Training Project for Regional Innovation (No. 2013H1B8A2027455)." REFERENCES: Akca, D., Gruen, A., 2009. "Comparative geometric and radiometric evaluation of mobile phone and still video cameras", The Photogrammetric Record, 24(127), pp.217-245. Fraser, C.S., 1997, "Digital camera self-calibration", ISPRS Journal of Photogrammetry and Remote Sensing, Vol.52, No.4, pp.149-159. Fraser, C.S., Al-Ajlouni, S., 2006, "Zoom-dependent camera calibration in digital close- range photogrammetry", Photogrammetric Engineering & Remote Sensing, Vol.72, No.9, pp.1017-1026. Habib, A., Morgan, M., 2003, "Automatic calibration of low-cost digital cameras", Journal of Optical Engineering, Vol.42, No.4, pp.948-955. Pullivelli, A.M., 2005, Low-cost digital cameras : calibration, stability analysis, and applications, University of Calgary, p.9.