Automatic target recognition by matching oriented edge pixels

Abstract

This paper describes techniques to perform efficient and accurate target recognition in difficult domains. In order to accurately model small, irregularly shaped targets, the target objects and images are represented by their edge maps, with a local orientation associated with each edge pixel. Three dimensional objects are modeled by a set of two-dimensional (2-D) views of the object. Translation, rotation, and scaling of the views are allowed to approximate full three-dimensional (3-D) motion of the object. A version of the Hausdorff measure that incorporates both location and orientation information is used to determine which positions of each object model are reported as possible target locations. These positions are determined efficiently through the examination of a hierarchical cell decomposition of the transformation space. This allows large volumes of the space to be pruned quickly. Additional techniques are used to decrease the computation time required by the method when matching is performed against a catalog of object models. The probability that this measure will yield a false alarm and efficient methods for estimating this probability at run time are considered in detail. This information can be used to maintain a low false alarm rate or to rank competing hypotheses based on their likelihood of being a false alarm. Finally, results of the system recognizing objects in infrared and intensity images are given.

Efficient algorithms for robust feature matching

“…These range from relaxation-based methods [19,4], to cluster detection in transformation space (by computing point-to-point correspondences [20][21][22], to hierarchical decomposition of transformation space coupled with the application of a robust similarity measure [2,11,23,24]. Most of the techniques presented in these papers are computationally intensive (in a worst-case theoretical sense), or take long times to run in practice.…”

Learning a Kinematic Prior for Tree-Based Filtering

“…Toyama and Blake [14] show how the quadratic chamfer function can be turned into a likelihood which is approximately Gaussian. Edge orientation is included by computing the distance only for edges with similar orientation, in order to make the distance function more robust [9]. We also exploit the fact that part of an edge normal on the interior of the contour should be skin-coloured, and only take those edges into account [8].…”

“…These methods are made efficient by the use of distance transforms to compute the chamfer or Hausdorff distance between template and image [2,6]. Multiple templates can be dealt with efficiently by building a tree of templates [3,9]. However, exhaustive search for the object is computationally expensive and results in jerky motion.…”

“…The search is made efficient by only evaluating sub-trees with high posterior probability, but care has to be taken when setting the thresholds for these decisions so as to not discard good hypotheses too early. Comparison to Tree-Based Detection A common way to build a tree of templates is to cluster templates using a similarity measure [3,9]. The idea is to group similar templates and represent them with a single prototype template together with an estimate of the variance of the error within the cluster, which is used to define a matching threshold.…”

Multi-class Vehicle Type Recognition System

“…An oriented-contours points matrix E I is obtained using an histogram based threshold process. Each edge point p i of E I is considered as a vector in 3 : p i =[x,y,o]', where (x,y) is the point position, and o is the gradient orientation of p i [11]. We sample the gradient orientations to N bins.…”