Analysis of Clustering Tools for Moving Target Indication

This processing approach detects and tracks slow-moving targets inside buildings.

Current and future forces operating in urban environments need the capability to detect slow-moving personnel inside buildings. To accomplish this, a time-domain approach was developed that uses a low-frequency, ultra-wideband (UWB) radar with a transmit pulse capable of penetrating through the wall. The UWB corresponds to a high-range resolution that gives the capability to better locate the moving target (MT). Previously, the effectiveness of a time-domain, moving target indication (MTI) approach was demonstrated for detecting moving personnel inside wood and cinderblock structures, moving personnel walking in nonlinear trajectories, and multiple moving personnel walking in linear trajectories.

A time-domain approach is suggested as an alternative to a frequency-domain approach such as Doppler processing, since a very small Doppler shift in backscattered frequency is generated due to the slow motion of the mover, and the low frequency needed to penetrate through the wall. The time-domain processing algorithms are based on the change detection (CD) paradigm, which is inherently similar to clutter cancellation. In the CD paradigm, the Synchronous Impulse Reconstructive (SIRE) radar remains stationary and generates a set of images for a region of interest (ROI). Each image in the set is formed every two-thirds of a second. The stationary objects in the building remain in the same location in each image; however, moving personnel will be at different locations. The moving personnel can be detected by subtracting adjacent images in the set, thereby eliminating the stationary objects and identifying the MT signature.

The algorithms in the MTI processing formulation can be implemented in a real-time or near-real-time system; however, a person-in-the-loop is needed to select input parameters for the k-Means clustering algorithm. Specifically, the number of clusters input into the k-Means routine is unknown and requires manual selection. A critical need exists to automate all aspects of the MTI processing formulation, which requires using an additional processing routine to identify the number of clusters input into the k-Means algorithm. Two techniques were investigated that automatically determine the number of clusters: the knee-point (KP) algorithm and the recursive pixel finding (RPF) algorithm. The KP algorithm is a well-known heuristic approach for determining the number of clusters. The RPF algorithm is analogous to the image processing, pixel labeling procedure. Both routines were used to process data collected by the low-frequency UWB radar and compared their results.

The KP and RPF algorithms provide the k-Means algorithm with the number of clusters present in the morphological image. A cluster is defined as a group of one or more connected POIs. Unlike the KP algorithm, the RPF algorithm does not implement the k-Means algorithm or rely on a heuristic to determine the number of clusters. Instead, the RPF algorithm scans the morphological image pixel-by-pixel for a POI. The POI is a binary (1 or 0) value and signifies that a cluster has been identified. The RPF algorithm then recursively scans every pixel of the cluster in a local area and identifies the cluster boundaries. This process is similar to a pixel labeling procedure in which the image is scanned pixel-by-pixel from left to right and top to bottom. A pixel is considered an “object” if the pixel equals 1; otherwise it is considered a “hole.” All neighboring pixels of an

object are examined and if a neighbor is an object, then it is merged with the object to which it is connected. The merged objects form a list that is used to identify all object pixels forming a single cluster.

Twelve data sequences of moving personnel were analyzed by the KP and RPF algorithms. This data consists of multiple scenarios (real time with no simulations) of personnel walking inside wood and cinderblock buildings. During data collection, the radar remains stationary and is positioned broadside to the wall and 38° of the broadside angle, which was chosen due to practical considerations. The 12 data sequences were also input into the RPF algorithm, which processed the 12 data sequences and correctly identified the number of clusters present in the morphological images.

One possible disadvantage of the RPF algorithm is that it does not have the capability to merge clusters in close proximity. This merge feature could be easily added to the RPF procedure, but would require a redefinition of what is considered to be a cluster.

Both the KP and RPF algorithms automate the MTI processing formulation and enable this formulation to be implemented in a real-time or near-real-time system. An advantage of the KP algorithm is its capability of merging clusters that are close. The advantages of the RPF algorithm are its capability to correctly identify the “true” number of clusters in all morphological images and its low computational complexity.

This work was done by Anthony Martone, Roberto Innocenti, and Kenneth Ranney of the Army Research Laboratory. For more information, download the Technical Support Package (free white paper) at www.defensetechbriefs.com/tsp  under the Information Sciences category. ARL-0100



This Brief includes a Technical Support Package (TSP).
Document cover
Analysis of Clustering Tools for Moving Target Indication

(reference ARL-0100) is currently available for download from the TSP library.

Don't have an account? Sign up here.