**5.2 Change detection and calculation of the 3D metrics**

The results of change detection using an integrated approach of SVM and together with DSM differencing are presented in **Figure 7**. **Figure 8** also shows the profile views of an excavated site in 2005 that was changed to a tall building in 2008. As seen in **Figures 6** and **7**, the level of noise is very low for the integrated approach compared with when DSM differencing only is used and the magnitude

**Figure 5.** *Vegetation classification.*

**63**

**6. Discussion**

**Figure 7.**

**Figure 8.**

*the new tall building.*

*New Metrics for Spatial and Temporal 3D Urban Form Sustainability Assessment Using Time…*

of change can be accessed for each pixel. Having these values for each pixel gives a major advantage for the calculation of 3D metrics as proposed in Section 4.2. While this method is recommended in this research, there are other problems which need to be resolved before developing it further. One problem is to differentiate between buildings and trees in the changed results. For the problem of occlusions among trees and building points, it seems that separation of these two classes of objects in the change detection result is very challenging. The approach of classifying buildings first and then determining change detection second, is also challenging as the

*Changes of profiles views for an excavation site to a new tall building from 2005 to 2008.(a) excavated site, (b)* 

*Results of 3D change detection between 2005 and 2008 from the integration of DSM differencing and SVM.*

The analysis of the changes of urban form including buildings and trees in 3D space is important because contributions of changes in tree canopy cover compared

differences of lidar point densities impact the results [29].

*DOI: http://dx.doi.org/10.5772/intechopen.89617*

**Figure 6.** *Building classification.*

*New Metrics for Spatial and Temporal 3D Urban Form Sustainability Assessment Using Time… DOI: http://dx.doi.org/10.5772/intechopen.89617*

**Figure 7.**

*Sustainability in Urban Planning and Design*

objectives is presented in the following sections:

**5.1 Building and vegetation classification from airborne lidar**

**5.2 Change detection and calculation of the 3D metrics**

and building classification, ground classification and creating a DSM. Each of these

For classification of buildings using ERDAS Imagine software, various thresholds are tested for the parameters of minimum slope, minimum area, plane offset, minimum height, maximum height and roughness and we came up with the optimum values to be 30, 100, 1, 2.5, 10,000 and 0.3 for each of the parameters, respectively (see **Figures 5** and **6**). The results show misclassifications between ground and buildings, building boundaries and vegetation. In addition, some of the buildings with multi-level attachments to the rooftops are classified as vegetation. These misclassifications can be overcome through postprocessing by careful manual adjustment.

The results of change detection using an integrated approach of SVM and together with DSM differencing are presented in **Figure 7**. **Figure 8** also shows the profile views of an excavated site in 2005 that was changed to a tall building in 2008. As seen in **Figures 6** and **7**, the level of noise is very low for the integrated approach compared with when DSM differencing only is used and the magnitude

**62**

**Figure 6.**

**Figure 5.**

*Vegetation classification.*

*Building classification.*

*Results of 3D change detection between 2005 and 2008 from the integration of DSM differencing and SVM.*

#### **Figure 8.**

*Changes of profiles views for an excavation site to a new tall building from 2005 to 2008.(a) excavated site, (b) the new tall building.*

of change can be accessed for each pixel. Having these values for each pixel gives a major advantage for the calculation of 3D metrics as proposed in Section 4.2. While this method is recommended in this research, there are other problems which need to be resolved before developing it further. One problem is to differentiate between buildings and trees in the changed results. For the problem of occlusions among trees and building points, it seems that separation of these two classes of objects in the change detection result is very challenging. The approach of classifying buildings first and then determining change detection second, is also challenging as the differences of lidar point densities impact the results [29].
