7 Common Pitfalls and Resolutions in Feature Extraction using TerraSolid and MicroStation

BAAM.Tech presents the first installment of survey grade geospatial tips and tricks. To continue education, bolster industry trends, and marvel at industry advancement, we are highlighting our best workflow practices and hurdles overcome. As an end-to-end production shop, our team is well-versed in industry knowledge and well-rounded in capabilities. From flight planning to data acquisition, post processing to final deliverables, our tools and team provide accurate information to enhance decision making. Throughout that process, we have continued to make strategic advances in our workflow. In an effort to push the boundary of feature extraction, we have learned a great deal. Here, 7 common pitfalls in feature extraction are covered including resolutions.

Our first step in post processing is to ensure our classified ground points are representative of the actual surface we flew. This involves both automatic and manual cleaning. Running our raw point cloud through the TerraScan drone wizard classifies our data (ground, low/medium/high vegetation, building, noise, etc.) and from there we can pan through the data for any anomalies. These are often points classified as ground that do not represent the real-world ground, such as at the bottom of drainage inlets or running up the edge of a building wall. These points will be manually classified out of the ground. The example below demonstrates an inlet in a parking lot where the bottom of the inlet was classified as ground. In the shaded surface, we can see the indentation in the surface while the cross-section view identifies erroneous ground points.

Classify these points as low points, to be deleted from the point cloud later, and we have a clean surface.

Low Points - Pre Classification
Low Points - Post Classification
Once you are in the groove of feature extraction and homed in collecting line after line after polygon, it is easy to overlook snaps to adjoining lines. This can be alleviated some by planning ahead when drawing edge of road surfaces, for example, and placing vertices where a sidewalk can be snapped in the future. Using the keypoint snap type, new lines will snap to the desired vertex and hold that XYZ value. Similarly, by using the nearest snap type, a line can be snapped anywhere along the adjacent feature. Towards the end of a project, these snaps (or lack of snaps) can be identified through the design file cleanup tool in MicroStation. In the example below, at the conclusion of drawing a sidewalk line, we turn on AccuSnap, set the type to Keypoint, and we can finish the line by snapping it to the corner of the paved surface feature.
AccuSnaps - Live Photo
Hanging lines overlapping polygons or other line features make for a messy looking deliverable. This is somewhat common for features that run into overlapping buildings. In the example below, a wall feature runs right up to a building polygon. Due to the elevation of the building roof, our polygon overhangs the wall feature. To clean the linework here and ensure our wall line is flush with the building, we will continue the wall line using the place orthogonal shape tool. At the conclusion of the shape, it is important to remember to go back and trim the over-extended wall to the edge of the building. To do so, select the building polygon, choose the Trim Multiple tool in MicroStation, and click the line to trim. It is important to recognize this practice works best for planimetric features as they will usually be delivered as 2D features. Conversely, when DTM features overlap, this trim tool will not create a vertex which could leave a gap in elevation between the features. A precise snap would be preferred in this case. We’ll get into differences between DTM and planimetrics next.
Untrimmed Feature
Trimming Shape with Label

One of the most common deliverables our clients ask for is a LandXML surface model. Generating a surface model requires DTM extraction in the form of breaklines. These are features that are on the ground and hold elevations for the area. Common features included in our DTM extraction are top and toe of slopes, edges of paved surfaces, road crown lines, gutters/top of curb, and edge of water. We will also include model keypoints or grid points to sub-sample the lidar ground points. For our simplified surfaces, we often use 50-foot grid points with a closest Z value sampling. These points are exported as an XYZ text file through TerraScan. Unneeded grid points (within 10 feet of breaklines, on roads or buildings) will be deleted. This lends itself to a better-looking (smoother) surface model in comparison to using model keypoints. These options are dependent on the level of accuracy required. Alternatively, planimetric features are not included in the surface model as they are only 2D and do not hold an elevation value. These features include anything else we can see in the project area – commonly tree lines, street signs, buildings, powerlines. Most of our planimetrics are extracted off the orthomosaic while the lidar can provide confirmation on location. Since these are representative 2D features, their elevation is set to 0 prior to delivery. Of course, this is client based and may differ from project to project. A following topic will key on this differentiation as well.

Export Lattice Model Pane
Rendered Lattice Model
Crossing DTM lines are another factor to look at when performing QC during or prior to completion of a project. When DTM lines are crossing, whether the lines were extracted using mouse point adjustment or draped to the ground, the elevation will likely not be constant at that intersection. The slightest difference in elevation between these lines will cause messy and inaccurate triangles in the surface model. For our process, we will normally break the DTM element at the crossing point (such as a breakline that crosses into riprap) as both these elements will hold the surface elevation. On the other hand, crossing DTM elements can be maintained if the lines are snapped to the same vertex at the crossing point. This ensures the same elevation at this point. In the accompanying image, breaklines are broken where they meet riprap. The shaded surface shows breakline locations on top and toe of slopes whereas the riprap area can be viewed in the ortho.
DTM - View 1
DTM - View 2
One of the trickier features to extract with consistency are horizontally and vertically offset lines such as edge of pavement, gutter line and top/back of curb. These lines in tandem are normally parallel horizontally with a consistent vertical offset between the gutter line and top of curb. Of course, this can vary depending on several factors such as types of curbs or type of road. Initial challenges for our team included top of curb lines that looked great from a top view, but we found they were at the same elevation as the gutter line when viewed as a cross section. Throughout an ongoing learning process, we have found success with a combination of TerraScan tools to create these DTM features as they were designed. This includes the Drape Linear Element tool (setting the Runs along option to Fixed height curb stone and Curb width/Curb height to 0.5 ft.), Move/Copy Parallel tool and Move Element tool. Using our ortho and lidar point cloud for extraction as well, we are on our way to overcoming this challenge and creating a consistently reproducible workflow.
Top View of Offset
Offset - RGB
Offset - Side View
Finally, with multiple people working on the same project simultaneously, consistency and communication is pivotal for a clean deliverable. Lack of communication could result in doubling up on feature extraction or inconsistency in line types/workspaces. Kick off meetings are a good way to get ahead of these issues to clarify client requirements for extraction. With multiple level libraries and cell libraries, getting on the same page avoids clashing linework and symbols. A final deliverable where it is indiscernible whether one person or a team completed the feature extraction is one of the most important goals. As an example, this screenshot shows three different power pole symbols, each from a different cell library.
Collaborative Extraction

Providing this brief look into our feature extraction processes, we hope you can gather some insights and extend the limits of your own practices. Reach out with any tips or tricks you would like us to dive deeper on or share any of your experiences.

Be on the lookout for our next installment!

Again, feel free to reach out with any questions or inquiries. Our team continues to evolve and innovate relating to our workflows and we would be happy to collaborate with you.

Recent Articles

The KCSI Team with the BAAM.Tech and AISPECO Teams infont of a plane carrying the AISPECO Heliux Lite Pod

KCSI Aerial Patrol Enhances Aerial Mapping Capabilities with AISPECO Helix Lite Pod from BAAM.Tech

BAAM.Tech partners with Inertial Labs

DJI Zenmuse L2 Accuracy Assessment and Use Cases by BAAM.Tech

The DJI L2 Accuracy Analysis and Use Case Exploration by BAAM.Tech

TERRACON, the premier North American Terrasolid networking and training event.
May 29 - 31, 2024 Kansas City, Mo
BAAM.Tech Logo

Shop by Category

Shop by Partner