Greenwich Institute for Science and Technology
   Home >> Modeling & Simulation >> Automated Feature Extraction (AFE) and Polyline Generation
Automated feature extraction (AFE) and polyline generation:

      The GIST techniques for rapidly identifying and extracting features from LIDAR data, remote sensing data, and other imagery are based on the functions of human vision system. For each image the entire set of objects is recognized as a single scene, areas of interest or specific objects are identified within that scene, and then those areas or objects are further interrogated for more detailed analysis. The human eye and brain operate on a similar principle. A human can see a large scene (including peripheral vision) and process all of the information rapidly to identify objects or areas of interest within that scene. Then the human brain focuses on those objects/areas to further identify more detailed features. As with the human brain its memory, the GIST methods use a composite attribution library, built from processing terabytes of LIDAR data, for classifying objects and features from the data. The GIST techniques emulate these processes and include the following elements:

  • Single scene processing
  • Fast one-dimensional saccades to search for objects
  • Fixation and effective 2-D extraction of a objects according specific criteria
  • Low-resolution vision (global or retinal vision) used for fast search of objects
  • High-resolution vision (local or fovea vision) used for accurate extraction of specific features
  • Feature and object attribution
Figure 1 illustrates the algorithm for processing LIDAR data with the bio-inspired feature extractor; the results of the extraction are seen in on Figure 2. The GIST techniques for processing LIDAR data and generating 3-D scenes operate at near real-time.

Scheme of algorithm

Fig. 1. The algorithm for processing LIDAR data, using the generation of polylines of roads as an example.


Fig. 2. Outputs from automated feature extraction. (Top). An extracted road depicted from connected 1.2-m resolution pixels and automatically generated polylines for centerlines with an optimized number of shape points and nodes. (Bottom). Polylines for a ship in Boston harbor.

2010 Greenwich Institute for Science and Technology