3.2.1. SmartSkeMa—Sketch Map Data Collection Software

SmartSkeMa is a software application that we developed to support the documentation of land tenure information for communities with a focus on the actual land practices in the communities. SmartSkeMa supports land recording processes in two main ways [13]. First, it provides a means to document land related concepts as expressed within the local culture or context in a structured domain model [45]. Second, it supports sketch-based community mapping processes by providing a means to digitize, annotate, and geolocalize hand-drawn objects in a sketch map [46]. The method uses both qualitative and quantitative representations of a digitized sketch map and aligns features from the sketch map with corresponding features in the base map. For qualitative representations alignment of qualitative spatial configurations is done. In the case of quantitative (cartesian) representations, the alignment is performed by a coordinate transformation using predetermined control points. The latter approach allows SmartSkeMa to be used as a digitizer over aerial imagery (Figure 5).

**Figure 5.** Screencast of a SmartSkeMa live demonstration of processing the spatial information drawing on top of a satellite image, the vector representation of drawn features as a SmartSkeMa response, and georeferencing drawn features (web-link: www.smartskema.eu).

Sketch maps are uploaded into SmartSkeMa as raster images. SmartSkeMa then converts these images into vector form in two steps. First any symbols in an image are detected and recognized using a Convolutional Neural Networks (CNN) trained on a set of predefined hand-drawn symbols. The symbols form a visual language for representing land use concepts and land features. After symbol detection the system performs a stroke-based image segmentation wherein boundaries of sketched objects are traced and extracted. Finally, the concepts corresponding to the detected symbols are applied to the image segments based on distance and a fixed set of rules specifying spatial constraints on configurations of different types of features.

The data collection used for the current study needed for the SmartSkeMa system was completed during a series of fieldworks and workshops with male and female members of the Massai community starting from 2017 in Kajiado county and Nairobi, Kenya and running through to October 2018. The sessions included demonstrations of the three main functional parts of the SmartSkeMa (Figure 6), followed by discussions about the applicability of SmartSkeMa. Questions were posed through questionnaires to evaluate the applicability of SmartSkeMa in (1) standard (official) land information recording processes, (2) documenting local land tenure systems, and (3) other land administration tasks.

**Figure 6.** Workflow of SmartSkeMa: Right side: local communities provide spatial and nonspatial information via sketch maps. Nonspatial information is processed via local domain model (LDM) and connected via the adapter model to land administration domain model (LADM). Spatial information is recognized via the object detection technique, captured qualitatively via the qualitative representations, and aligned with existing dataset such as feature extracted from UAV data.
