Semi-Automatic Approach for Optical and Lidar Data Integration Using Phase Congruency Model at Multiple Resolutions

Abstract

Abstract. In light of the ongoing urban sprawl reported in recent studies, accurate urban mapping is essential for assessing current status and evolve new policies, to overcome various social, environmental, and economic consequence. Imagery and LiDAR data integration densifies remotely sensed data with radiometric and geometric characteristics, respectively, for a precise segregation of different urban features. This study integrated aerial and LiDAR images using point primitives, which were obtained from running the Phase Congruency model as an image filter to detect edges and corner. The main objective is to study the effect of applying the filter at different spatial resolutions on the registration accuracy and processing time. The detected edge/corner points that are mutual in both datasets, were identified as candidate points. The Shape Context Descriptor method paired-up candidate points as final points based on a minimum correlation of 95%. Affine, second and third order polynomials, in addition to the Direct Linear Transformation models were applied for the image registration process using the two sets of final points. The models were solved using Least Squares adjustments, and validated by a set of 55 checkpoints. It was observed that with the decrease in spatial resolution, on one hand, the registration accuracy did not significantly vary. However, the consistency of the model development and model validation accuracies were enhanced, especially with the third order polynomial model. On the other hand, the number of candidate points decreased; consequently, the processing time significantly declined. The 3D LiDAR points were visualised based on the Red, Green, and Blue radiometric values that were inherited from the aerial photo. The qualitative inspection was very satisfactory, especially when examining the scene’s tiny details. In spite of the interactivity in determining the candidate points, the proposed procedure overcomes the dissimilarity between datasets in terms of acquisition technique and time, and widens the tolerance of accepting points as candidates by including points that are not traditionally considered (i.e. road intersections).

Fusion of Airborne LiDAR Point Clouds and Aerial Images for Heterogeneous Land-Use Urban Mapping

“…Early classification experiments revealed errors in the geo-registration performed in [45] using the third-order polynomial model. Figure 11 illustrates some of these errors featured by the displacement of sidewalks (Figure 11a) as well as misalignment of buildings and grasses (Figure 11c).…”

“…As concluded in [45,46], it is better to apply the PC model for LiDAR and imagery data geo-registration on 2D images resampled at lower spatial resolutions to avoid computationally expensive data processing and high model development error values. However, including CED with the PC filter as a scene abstraction approach significantly reduces the model development error values (from 5.74 to 1.78 pixels using the affine model) when processing the data at high spatial resolutions (40 cm).…”

“…A model's ability to minimize its development and validation errors should not be the sole criterion for model selection. For example, the third-order polynomial model produced the least and closest development and validation errors [45]; however, the model caused overparameterization that resulted in displacement and misalignment errors. Therefore, examining feature edges (i.e., building roofs and grasses) after ground filtering as well as a preliminary classification of some ground objects and matching them to a basemap are recommended practices for a reliable qualitative geo-registration assessment.…”

“…Despite the apparent satisfactory qualitative and quantitative validation of the georegistration results conducted in [45], initial classification trials showed misalignment of the LiDAR point clouds. The misalignment stemmed from the incorrect inheritance of radiometric properties from the aerial photo to the point clouds.…”

“…The LiDAR point clouds were geo-registered to an aerial image of the same study region in [45] to inherit additional spectral properties. The geo-registration approach was applied the PC (Phase Congruency) model and a scene abstraction approach to detect edge points in both aerial and LiDAR images.…”