Baseline processing pipeline for fast automatic target detection and recognition in airborne 3D ladar imagery

It has been proven that 3D ladar imagery has a strong potential for automatic target detection (ATD) and automatic target recognition (ATR); ladars enhance target information, which may then be exploited to yield higher recognition rates and lower false alarms. Although numerous techniques have been proposed for both 3D ATD and 3D ATR, no single approach has proven capable of systematically outperforming all other techniques for every possible scenario. In this context, this paper describes a set of fast 3D ATD/ATR algorithms designed to process cooperative targets in airborne 3D ladar imagery. This algorithmic chain consists of four modules: detection, segmentation, classification and recognition. In each module, fast algorithms were implemented, some of which stem from open literature while others were designed in-house. The purpose of this algorithmic chain is to provide a baseline approach for efficient processing of simple scenarios. The ultimate goal of this work is to characterize and compare algorithms with respect to increasingly complex scenarios, in hopes of progressing towards an adaptive processing pipeline for context-driven 3D ATD/ATR. In this paper, the four modules of the baseline processing pipeline are first described. Preliminary test results obtained with real airborne ladar imagery are then presented, in which fast and accurate 3D ATD/ATR is performed with a library of 20 scanned vehicles. Finally, a demonstration is presented to illustrate how this baseline approach may be expanded to tackle more complex scenarios, such as non-cooperative targets concealed under vegetation.