A Terrain Relative Navigation sensor enabled by multi-core processing

Survey on advances on terrain based navigation for autonomous underwater vehicles

“…A recent description of a commercial TBN system for aircraft has been done by Cowie et al (2008). More recently, TBN has also been proposed to be used on several space exploration missions, namely for planetary entry, descent and landing by Alexander et al (2012) and Johnson and Montgomery (2008). For a recent review on the different TBN solutions for aerial vehicles, refer to works by Karabork (2010); Vaman (2012) and the references therein.…”

Vision‐based, terrain‐aided navigation with decentralized fusion and finite set statistics

“…Another related tool is that of APLNav described by Adams et al that stores a map of the reference surface and renders “expected images” for correlation and comparison to those collected by the mobile sensor. The work presented by Alexander et al describes the many tools employed by these sophisticated image processing techniques, including but not limited to (a) map rendering and shading; (b) image normalization, shifting, and scaling; (c) map and image rectification; (d) the fast Fourier transform (FFT); and (e) spatial correlation of the image with the reference map. In addition, another theme present in these types of terrain aiding procedures, referred to as “image‐correlation” methods in this work, is the use of a past (stochastic) navigation solution to construct the “measurement” provided by the sensor, either as a warm start for the map‐to‐image correlation methods or to construct an “expected image” of the environment.…”

“…This work chooses to shift the computational burden from advanced image processing to new developments in state‐of‐the‐art multitarget filtering techniques, rather than forcing a data type into a format that can be crudely interpreted as a traditionally modeled measurement type. The key to this approach is to track the features in collected images as “targets” without requiring knowledge of their identity and, in lieu of correlating them to a known reference map via spatial correlation of rendered and expected images such as in previous studies, instead using the “interesting” data within the images directly as features, such as making use of Harris corners, scale‐invariant feature transform (SIFT) features, or speeded up robust features (SURF) . The resulting navigation scheme sheds most of the advanced image processing in favor of a well‐modeled data type without any required reliance on an a priori reference map and, indeed, returns to treating the landing navigation problem as a SLAM problem.…”

“…At first, a navigator is tempted to manipulate the incoming measurements to conform with traditional filtering methods, such as the described image‐correlation methods. However, even if one is willing to assume that the camera and image processing are perfect, this approach presents a number of challenges/limitations to a standard navigation approach: Most approaches require the use of a known reference map to which captured images are correlated The approaches that utilize features that are unknown by the filter a priori utilize sequential images to reduce growth in uncertainty rather than processing the image contents to reduce state uncertainty itself …”

“…The approaches that utilize features that are unknown by the filter a priori utilize sequential images to reduce growth in uncertainty rather than processing the image contents to reduce state uncertainty itself Without an explicit model for the image processing, there is no way to rigorously quantify and estimate correlation errors. Construction of the “expected image” is often initialized with the previous navigation state as a warm‐start procedure, making the measurement statistically dependent on the previous navigation state If one seeks to estimate these map errors in an EKF‐like filter, broad assumptions must often be made on the errors (such as requiring that all feature errors are caused by a common map‐tie error). These approaches limit the vehicle's exploratory authority to remain within the confines of the pre‐loaded reference map …”