performance evaluation of lightweight lidar for uav ... - Semantic Scholar
PERFORMANCE EVALUATION OF LIGHTWEIGHT LIDAR FOR UAV APPLICATIONS Salvatore Espositoa, Matteo Murab, Paolo Fallavollitaa, Marco Balsia, Gherardo Chiricib, Arturo Oradinic, Marco Marchettib a
S. Esposito, P. Fallavollita, and M. Balsi are with the Department of Information Engineering, Electronics, and Telecommunications of “La Sapienza” University of Rome, Italy and Oben srl, Sassari, Italy (http://www.oben.it) b M. Mura, G. Chirici, and M. Marchetti are with the Department of Bioscience and Territory of University of Molise, Italy c A. Oradini is with Forestlab Centre S.r.l., Firenze, Italy Corresponding author is S. Esposito: +39-328-5950642; fax: +39-06-44585-918; e-mail: [email protected] 1. INTRODUCTION Airborne Laser Scanning (ALS) is increasingly used in support of forest resources inventory and monitoring, [1] – [4], but high cost of the surveys, especially for small surveyed areas [5], limit applicability. In particular, multitemporal surveys, which are of relevant importance in monitoring forest ecosystems at the small scale level, are in many cases not affordable. The largest part of the cost is due to renting the aircraft with its crew, be it either helicopter or airplane of different sizes. Recent progress in remote and autonomous control technology led recently to rapid growth of use of Unmanned Aerial Vehicles (UAVs) to carry active or passive sensors for aerial surveys, especially in small-area cases. A non-exhaustive list of UAV applications in survey includes agriculture [6] land cover [7], rapid terrain mapping [8]. In the LiDAR field the capability of UAVs to fly at lower altitudes and slower speed than man-driven aircraft results in higher point density, which is quite important for forestry applications [9], however large weight and size of most LiDAR sensors prevents their use with small UAVs, governed by less restrictive regulation, and requiring very simple and cheap logistics. For these reasons, in this work we assess the performance of a very light and small LiDAR [10], especially appropriate to be carried on a UAV, for this feasibility study mounted on an ultralight helicopter. The data derived from the survey has been subject to qualitative and quantitative assessment through a comparison with ground-truth data. 2. MATERIALS AND METHODS 2.1 Study area The study area, fig. 1, is a permanent plot of the University of Molise located in the Natural Reserve of Collemeluccio-Montedimezzo, situated in the center-North of Molise region, Italy, near the border with Abruzzo region. Since 1971 (DM 11/09/1971) the Reserve is part of the UNESCO Man and Biosphere (MaB) program, that aims at harmonizing the relationship between people and their surrounding natural environment. It is also a
Site of Community Importance (SIC, directive 92/43/CEE) and Special Protection Zone (ZPS, directive 79/409/CEE). The reserve extends for 291 ha, with an altitude range varying from 822 to 1284 m a.s.l. Along this altitudinal gradient, Turkey oak forests (Quercus cerris) dominate the lower altitudes and European beech forest (Fagus sylvatica) the higher, with frequent mixed situations according to topography, soil conditions and past management regimes. Several species of Maple (Acer sp.), Hornbeam (Carpinus sp.) and Ash (Fraxinus sp.) enrich the Reserve biodiversity. The area considered for this case study is 1km2, fig. 1, with special focus on a square plot of 1 ha area (side 100m), monitored by the University of Molise. To assess the ALS performance we used ground data collected during the field surveys carried out in 2008 in the framework of the project "Innovative methods for the identification, characterization and management of old-growth forests in the Mediterranean environment". 2.2 ALS Data The scan was performed using a YellowScan LiDAR [10], mounted on an ultralight helicopter. Yellowscan is integrated with internal INS capability and GPS receiver. GPS enhancement by RTK was planned, but did not succeed. For this reason, georeferencing accuracy was limited by GPS precision. Dimensions, weight and power consumption of this LiDAR are 15×15×15cm3, 2kg, and 20W respectively. Such characteristics make this LiDAR a good solution for light UAV applications. YellowScan operates up to 150 m above ground level. It provides high density measurements with an accuracy of 30 cm. The typical scan angle range is ±25° and can be increased up to ±50°. The system provides up to 3 echoes per shot, allowing to get topographic information under vegetation cover. Data were taken in a single flight, on Oct. 4, 2013. The raw ALS data were elaborated using LAStools software, [12]. Scanned area was composed of ten strips, labeled as L0-L9. In fig. 1, the strips are shown; in particular the strip which doesn’t show overlapping with other strips is L0. For every strip we computed the total points density, ground-only points density and the ground points density referred to the echo return, as shown in Table 1. All strips show a total density greater than 10 pts/m2, the minimum and maximum of the total density are obtained for the strips L0 (11.27 pts/m2) and L3 (63.65 pts/m2) respectively. It is also important to note that in all strips we also obtained second and third returns, classified as ground. The strip labeled as “merge” was obtained by merging strips L1 to L9. After removing noise points (outliers), the points were classified as ground / not ground, then the normalized height and the TIN interpolation of the raster surfaces (DTM and DSM), for L0 and merge strips, were computed. We also computed the canopy height model (CHM), as difference of the heights estimated by digital surface model (DSM) and digital terrain model (DTM).
fig. 1. Localization of the strips. The bigger red square defines the area under study The smaller red square defines the main survey target. Table 1. Total points density (pts/m2), ground-only points density (pts/m2), and ground points density referred to the echo return (1st, 2nd, 3rd). The strip labeled as merge is obtained by merging strips L1 to L9. Density Ground Ground_1st Ground_2nd Ground_3rd Strip Total Density Density Density Density L0 L1 L2 L3 L4 L5 L6 L7 L8 L9 Merge (L1-L9)
11.27 52.67 26.3 63.65 28.78 38.01 51.39 25.57 25.57 31.44 49.01
0.65 0.84 0.69 0.92 0.68 0.78 0.75 0.59 0.6 0.86 0.78
0.51 0.46 0.38 0.45 0.39 0.53 0.49 0.7 0.62 0.49 0.53
0.51 0.61 0.58 0.68 0.53 0.62 0.58 0.45 0.48 0.68 0.59
0.36 0.58 0.43 0.62 0.46 0.52 0.49 0.37 0.37 0.52 0.51
3. RESULTS Processing the ALS data labelled as “merge” in the previous section, we were able to estimate DTM and the mean height of the trees respectively. In particular in order to check goodness of the estimated DTM we have compared it with the one interpolated from the contours of the local technical regional map (CTR) at scale 1:10,000 (DTM_CTR). Applying linear regression to these two datasets, we obtained good fitting of the ALS data, in fact the coefficient of determination (R2) is close to unity. Although the angular coefficient of the regression line was close to unity the intercept was not equal to zero. This was a systematic error probably due to geo-referring error (also taking into account that the plot is steeply sloped, so that small displacement causes large height error). In the same way we compared the tree height estimated by CHM with ground truth data. In this case we were able to estimate the mean height of the trees from the estimated CHM with relative error equal to 5%. In fact, the mean height measured on the ground and measured from estimated CHM were 27.55m and 29.02m respectively. Due to
geo-referring error of the individual trees (both as scanned, and as surveyed on ground) it was not possible to define a linear relation between the two datasets, but the overall agreement of the estimated CHM with ground truth is remarkable. 4. CONCLUSIONS Based on our considerations on these preliminary results, we believe that the Yellowscan LiDAR can be used for forestry application. In fact, using the ALS data we were able to estimate DTM with error consistent with GPS errors, and the mean height with relative error within 5%. This preliminary results show that the YellowScan LiDAR is able to estimate forest parameters in area-based approach. Better georeferencing is necessary to improve quality of the surveys: use of differential and/or multi-constellation GNSS receivers could enhance the result substantially. Yellowscan, being the lightest LiDAR available on the market at the time of the survey, is suited for light UAVs complying with simplified regulations for UAVs weighing less than 25kg, already in force or being drafted in EU countries. 5. REFERENCES [1] Corona, P., Cartisano, R., Salvati, R., Chirici, G., Floris, A., di Martino, P., Torresan, C. Airborne laser scanning to support forest resource management under alpine, temperate and Mediterranean environments in Italy. Italian Journal of Remote Sensing / Rivista Italiana di Telerilevamento, 45(1), 27–37, 2012 [2] Lim, K., Treitz, P., Wulder, M., St-Ongé, B., & Flood, M. LiDAR remote sensing of forest structure. Progress in Physical Geography, 27(1), 88–106, 2003 [3] Montaghi, A., Corona, P., Dalponte, M., Gianelle, D., Chirici, G., & Olsson, H. Airborne laser scanning of forest resources: An overview of research in Italy as a commentary case study. International Journal of Applied Earth Observation and Geoinformation, Vol. 23, Pages 288–300, 2013 [4] Wulder, M. A., Bater, C. W., Coops, N. C., Hilker, T., & White, J. C. The role of LiDAR in sustainable forest management. Forestry Chronicle, Vol 84(6), 807–826, 2008. [5] Hummel, S., Hudak, A. T., Uebler, E. H., Falkowski, M. J., & Megown, K. A. A comparison of accuracy and cost of Lidar versus stand exam data for landscape management on the Malheur national forest. Journal of Forestry, Vol. 109(5), 267–273, 2011. [6] Hunt, E. R., Hively, W. D., McCarty, G. W., Daughtry, C. S. T., Forrestal, P. J., Kratochvil, R. J., Miller, C. D. NIR-Green-Blue highresolution digital images for assessment of winter cover crop biomass. GIScience & Remote Sensing, 48(1), 86–98. doi:10.2747/15481603.48.1.86, 2011 [7] Tenenbaum, D. E., Yang, Y., & Zhou, W. A comparison of object-oriented image classification and transect sampling methods for obtaining land cover information from digital orthophotography. GIScience & Remote Sensing, 48(1), 112–129. doi:10.2747/15481603.48.1.112, 2011. [8] Stefanik, K. V, Gassaway, J. C., Kochersberger, K., & Abbott, A. L. UAV-based stereo vision for rapid aerial terrain mapping. GIScience & Remote Sensing, 48(1), 24–49. doi:10.2747/1548-1603.48.1.24, 2011. [9] Jaakkola, A., Hyyppä, J., Kukko, A., Yu, X., Kaartinen, H., Lehtomäki, M., & Lin, Y. (2010). A low-cost multi-sensoral mobile mapping system and its feasibility for tree measurements. Isprs Journal of Photogrammetry and Remote Sensing, Vol 65(6), 514–522. [10] LiDAR Website, http://yellowscan.lavionjaune.com/, last access Dec, 2013 [11] LAStools Website, http://rapidlasso.com/lastools/, last access Dec, 2013
S. Esposito, P. Fallavollita, and M. Balsi are with the Department of Information Engineering, Electronics, and Telecommunications of “La Sapienza” University of Rome, Italy and Oben srl, Sassari, Italy (http://www.oben.it) b M. Mura, G. Chirici, and M. Marchetti are with the Department of Bioscience and Territory of University of Molise, Italy c A. Oradini is with Forestlab Centre S.r.l., Firenze, Italy Corresponding author is S. Esposito: +39-328-5950642; fax: +39-06-44585-918; e-mail: [email protected] 1. INTRODUCTION Airborne Laser Scanning (ALS) is increasingly used in support of forest resources inventory and monitoring, [1] – [4], but high cost of the surveys, especially for small surveyed areas [5], limit applicability. In particular, multitemporal surveys, which are of relevant importance in monitoring forest ecosystems at the small scale level, are in many cases not affordable. The largest part of the cost is due to renting the aircraft with its crew, be it either helicopter or airplane of different sizes. Recent progress in remote and autonomous control technology led recently to rapid growth of use of Unmanned Aerial Vehicles (UAVs) to carry active or passive sensors for aerial surveys, especially in small-area cases. A non-exhaustive list of UAV applications in survey includes agriculture [6] land cover [7], rapid terrain mapping [8]. In the LiDAR field the capability of UAVs to fly at lower altitudes and slower speed than man-driven aircraft results in higher point density, which is quite important for forestry applications [9], however large weight and size of most LiDAR sensors prevents their use with small UAVs, governed by less restrictive regulation, and requiring very simple and cheap logistics. For these reasons, in this work we assess the performance of a very light and small LiDAR [10], especially appropriate to be carried on a UAV, for this feasibility study mounted on an ultralight helicopter. The data derived from the survey has been subject to qualitative and quantitative assessment through a comparison with ground-truth data. 2. MATERIALS AND METHODS 2.1 Study area The study area, fig. 1, is a permanent plot of the University of Molise located in the Natural Reserve of Collemeluccio-Montedimezzo, situated in the center-North of Molise region, Italy, near the border with Abruzzo region. Since 1971 (DM 11/09/1971) the Reserve is part of the UNESCO Man and Biosphere (MaB) program, that aims at harmonizing the relationship between people and their surrounding natural environment. It is also a
Site of Community Importance (SIC, directive 92/43/CEE) and Special Protection Zone (ZPS, directive 79/409/CEE). The reserve extends for 291 ha, with an altitude range varying from 822 to 1284 m a.s.l. Along this altitudinal gradient, Turkey oak forests (Quercus cerris) dominate the lower altitudes and European beech forest (Fagus sylvatica) the higher, with frequent mixed situations according to topography, soil conditions and past management regimes. Several species of Maple (Acer sp.), Hornbeam (Carpinus sp.) and Ash (Fraxinus sp.) enrich the Reserve biodiversity. The area considered for this case study is 1km2, fig. 1, with special focus on a square plot of 1 ha area (side 100m), monitored by the University of Molise. To assess the ALS performance we used ground data collected during the field surveys carried out in 2008 in the framework of the project "Innovative methods for the identification, characterization and management of old-growth forests in the Mediterranean environment". 2.2 ALS Data The scan was performed using a YellowScan LiDAR [10], mounted on an ultralight helicopter. Yellowscan is integrated with internal INS capability and GPS receiver. GPS enhancement by RTK was planned, but did not succeed. For this reason, georeferencing accuracy was limited by GPS precision. Dimensions, weight and power consumption of this LiDAR are 15×15×15cm3, 2kg, and 20W respectively. Such characteristics make this LiDAR a good solution for light UAV applications. YellowScan operates up to 150 m above ground level. It provides high density measurements with an accuracy of 30 cm. The typical scan angle range is ±25° and can be increased up to ±50°. The system provides up to 3 echoes per shot, allowing to get topographic information under vegetation cover. Data were taken in a single flight, on Oct. 4, 2013. The raw ALS data were elaborated using LAStools software, [12]. Scanned area was composed of ten strips, labeled as L0-L9. In fig. 1, the strips are shown; in particular the strip which doesn’t show overlapping with other strips is L0. For every strip we computed the total points density, ground-only points density and the ground points density referred to the echo return, as shown in Table 1. All strips show a total density greater than 10 pts/m2, the minimum and maximum of the total density are obtained for the strips L0 (11.27 pts/m2) and L3 (63.65 pts/m2) respectively. It is also important to note that in all strips we also obtained second and third returns, classified as ground. The strip labeled as “merge” was obtained by merging strips L1 to L9. After removing noise points (outliers), the points were classified as ground / not ground, then the normalized height and the TIN interpolation of the raster surfaces (DTM and DSM), for L0 and merge strips, were computed. We also computed the canopy height model (CHM), as difference of the heights estimated by digital surface model (DSM) and digital terrain model (DTM).
fig. 1. Localization of the strips. The bigger red square defines the area under study The smaller red square defines the main survey target. Table 1. Total points density (pts/m2), ground-only points density (pts/m2), and ground points density referred to the echo return (1st, 2nd, 3rd). The strip labeled as merge is obtained by merging strips L1 to L9. Density Ground Ground_1st Ground_2nd Ground_3rd Strip Total Density Density Density Density L0 L1 L2 L3 L4 L5 L6 L7 L8 L9 Merge (L1-L9)
11.27 52.67 26.3 63.65 28.78 38.01 51.39 25.57 25.57 31.44 49.01
0.65 0.84 0.69 0.92 0.68 0.78 0.75 0.59 0.6 0.86 0.78
0.51 0.46 0.38 0.45 0.39 0.53 0.49 0.7 0.62 0.49 0.53
0.51 0.61 0.58 0.68 0.53 0.62 0.58 0.45 0.48 0.68 0.59
0.36 0.58 0.43 0.62 0.46 0.52 0.49 0.37 0.37 0.52 0.51
3. RESULTS Processing the ALS data labelled as “merge” in the previous section, we were able to estimate DTM and the mean height of the trees respectively. In particular in order to check goodness of the estimated DTM we have compared it with the one interpolated from the contours of the local technical regional map (CTR) at scale 1:10,000 (DTM_CTR). Applying linear regression to these two datasets, we obtained good fitting of the ALS data, in fact the coefficient of determination (R2) is close to unity. Although the angular coefficient of the regression line was close to unity the intercept was not equal to zero. This was a systematic error probably due to geo-referring error (also taking into account that the plot is steeply sloped, so that small displacement causes large height error). In the same way we compared the tree height estimated by CHM with ground truth data. In this case we were able to estimate the mean height of the trees from the estimated CHM with relative error equal to 5%. In fact, the mean height measured on the ground and measured from estimated CHM were 27.55m and 29.02m respectively. Due to
geo-referring error of the individual trees (both as scanned, and as surveyed on ground) it was not possible to define a linear relation between the two datasets, but the overall agreement of the estimated CHM with ground truth is remarkable. 4. CONCLUSIONS Based on our considerations on these preliminary results, we believe that the Yellowscan LiDAR can be used for forestry application. In fact, using the ALS data we were able to estimate DTM with error consistent with GPS errors, and the mean height with relative error within 5%. This preliminary results show that the YellowScan LiDAR is able to estimate forest parameters in area-based approach. Better georeferencing is necessary to improve quality of the surveys: use of differential and/or multi-constellation GNSS receivers could enhance the result substantially. Yellowscan, being the lightest LiDAR available on the market at the time of the survey, is suited for light UAVs complying with simplified regulations for UAVs weighing less than 25kg, already in force or being drafted in EU countries. 5. REFERENCES [1] Corona, P., Cartisano, R., Salvati, R., Chirici, G., Floris, A., di Martino, P., Torresan, C. Airborne laser scanning to support forest resource management under alpine, temperate and Mediterranean environments in Italy. Italian Journal of Remote Sensing / Rivista Italiana di Telerilevamento, 45(1), 27–37, 2012 [2] Lim, K., Treitz, P., Wulder, M., St-Ongé, B., & Flood, M. LiDAR remote sensing of forest structure. Progress in Physical Geography, 27(1), 88–106, 2003 [3] Montaghi, A., Corona, P., Dalponte, M., Gianelle, D., Chirici, G., & Olsson, H. Airborne laser scanning of forest resources: An overview of research in Italy as a commentary case study. International Journal of Applied Earth Observation and Geoinformation, Vol. 23, Pages 288–300, 2013 [4] Wulder, M. A., Bater, C. W., Coops, N. C., Hilker, T., & White, J. C. The role of LiDAR in sustainable forest management. Forestry Chronicle, Vol 84(6), 807–826, 2008. [5] Hummel, S., Hudak, A. T., Uebler, E. H., Falkowski, M. J., & Megown, K. A. A comparison of accuracy and cost of Lidar versus stand exam data for landscape management on the Malheur national forest. Journal of Forestry, Vol. 109(5), 267–273, 2011. [6] Hunt, E. R., Hively, W. D., McCarty, G. W., Daughtry, C. S. T., Forrestal, P. J., Kratochvil, R. J., Miller, C. D. NIR-Green-Blue highresolution digital images for assessment of winter cover crop biomass. GIScience & Remote Sensing, 48(1), 86–98. doi:10.2747/15481603.48.1.86, 2011 [7] Tenenbaum, D. E., Yang, Y., & Zhou, W. A comparison of object-oriented image classification and transect sampling methods for obtaining land cover information from digital orthophotography. GIScience & Remote Sensing, 48(1), 112–129. doi:10.2747/15481603.48.1.112, 2011. [8] Stefanik, K. V, Gassaway, J. C., Kochersberger, K., & Abbott, A. L. UAV-based stereo vision for rapid aerial terrain mapping. GIScience & Remote Sensing, 48(1), 24–49. doi:10.2747/1548-1603.48.1.24, 2011. [9] Jaakkola, A., Hyyppä, J., Kukko, A., Yu, X., Kaartinen, H., Lehtomäki, M., & Lin, Y. (2010). A low-cost multi-sensoral mobile mapping system and its feasibility for tree measurements. Isprs Journal of Photogrammetry and Remote Sensing, Vol 65(6), 514–522. [10] LiDAR Website, http://yellowscan.lavionjaune.com/, last access Dec, 2013 [11] LAStools Website, http://rapidlasso.com/lastools/, last access Dec, 2013
