LiDAR point classification, quite simply, is the process of organizing points of similar type or characteristic into groups. In the LAS specifications, developed by the American Society for Photogrammetry & Remote Sensing (ASPRS), base classifications are provided with values and general meanings. As you read through the meanings, you quickly notice that the classifications are quite general. Unlike biology, where you have detailed breakdowns to classify specific organisms (family, genus, species), LiDAR technicians are provided generalized buckets with high-level definitions, with the added ability to input user defined groups. Another noticeable item is that the classifications are aerial LiDAR centric, which have evolved through versions just as the technology and applications have from primarily a terrain mapping tool, to applications in transmission corridors and transportation.
Often when dealing with Mobile LiDAR data that involves an urban environment, it becomes necessary to derive as much information as possible from the point cloud. Mobile mapping platforms generate a very feature-rich data set which is much higher resolution than those created with aerial systems that include many distinguishable features that would normally be linked to a GIS database. These features can include, but are not limited to, street signs, manholes, power lines, curb and gutter, storm inlets, light poles, and buildings among others.
There are many ways to utilize or mine information contained in the point cloud, but regardless of end product or employed techniques, it is always good practice to classify the cloud at the outset, so that downstream extraction efficiencies can be achieved. The user can define many layers or levels to parse classified points from the cloud. It is quite common during Mobile LiDAR collections to have erroneous laser returns generated by the system. These are rarely, if ever, widespread or prolific throughout the dataset, but its always good practice to first eliminate these from the data set by using automated routines followed by a manual inspection, prior to commencing detailed feature classifications.
Image Caption – A classified point cloud of an urban environment shows buildings, trees, road surfaces, signs and utilities. The points are shown with a DTM and contours.
Unless youve found the silver bullet application that automates the classification of the entire point-cloud correctly by clicking a single button (or more aptly waiving a magic wand), you have to start classifying points somewhereand theres usually no better place to begin than with road or street segments, especially if you dont have an automated routine, as these are usually readily identifiable. Assuming you dont have automated routines in your workflow to auto-classify, its typically easiest to draw a polygon using a top view of the data to classify road edges and curb-and-gutters, whereas the utilization of narrow cross sections in a profile view are customarily more effective to define more specific 3-dimensional features (i.e. light poles, sign posts, fire hydrants, etc). This process can be used in the extraction of many elements contained within the LiDAR data; however, it should be noted that this can be time consuming, so understanding your needs becomes critical for an efficient workflow and turnaround.
The downstream benefits to data classification can be quite substantial, especially if you have the pre-knowledge that the point-cloud will be re-purposed for other projects or applications. Classified data presents an easy filtering mechanism to reduce unnecessary noise in the extraction environment; effectively enabling technicians to target their focus on single elements, and more rapidly extracting those objects. Users can also easily benefit by reductions in data volume. Smaller data sets can also yield access to an expanded workforce whose computers might previously been unable to handle the larger volumes of data. An easy way to think of it is that classification equals downstream speed and efficiency.
Given the proliferation of Mobile LiDAR systems, the technology may present unique opportunities to yet again, revisit LAS specifications, and more specifically, classification. The feature rich data collection from a mobile unit captures endless variations of features that can be classified. Recently, a scope of work I reviewed required classified LAS data. Similar to the early requirement of full planimetrics, these specifications require discussion as to intent and extent. So by understanding the needs of the client, point classification is a very good method of delivering just the critical information that is desired. This helps to maximize the benefits of Mobile LiDAR data and minimize the impact from huge data volumes.