SPx Server, a component of the SPx radar processing family, is a COTS primary radar data extractor and target tracker. Interfacing to hardware or network radar video, SPx Server accepts polar format radar video and processes it to identify targets, which are then correlated from scan to scan to output positional and motion updates. The software is highly configurable and may be used to identify target-like shapes according to defined rules. These candidate tracks may be output directly after detection, for example into an existing correlator, or may be further processed by SPx Server using a track filter.

The SPx Server software module is integrated with the SPx Processing library, which provides a comprehensive range of radar processing capabilities. After acquisition and prior to data processing, the video may be passed through the standard SPx processing functions to implement gain control, area-based video removal, thresholding, interference suppression or other functions. Additionally, user-defined processes may be incorporated into the processing chain to allow a completely custom solution using SPx as the integrating framework.

When extracted data is correlated from scan to scan, the track correlator uses multiple hypotheses to support ambiguous interpretations of the radar video. The filter uses position, size, shape and historical measurements to correlate existing tracks with new data, providing updated positions and dynamics, as well as a confidence estimate. The behaviour of historical track data is analysed to help interpretation and provide a first-level classification capability.

Optional video recording capabilities are available in the compatible SPx Record module that records the polar radar video to support, for example, incident recording or training applications.

Target Extraction

The SPx Extraction process examines the processed video to search for target-like returns that form a connected target-like shape. A set of configurable parameters define the target size of interest, allowing small noise returns or larger clutter or land masses to be eliminated early in the processing.

The Extraction process begins by creating a set of spans that represent intervals of video above a threshold for each processed return. These spans are then combined across returns to form connected two-dimensional shapes. The weighted centre of gravity, bounding box and total weight of the plot shape are calculated and entered into a plot database, along with a timestamp. At this stage of the processing, no merging occurs of close plots that are likely to be derived from the same target. This allows partial plots to be reported on the network, if desired, and allows the tracking process to consider the merits of merging in the context of the local tracks.

Track Creation

The tracker maintains an active track database. The contents of the database are updated with new plot data derived from the data extraction stage. New tracks are added to the database from either a manual request (perhaps derived from an operator or else an external process), or else automatically.

The automatic track creation occurs when plots entered into the database are seen to be uncorrelated, or ungated, with any existing known target. A new preliminary track is created and is updated with future detections until confidence is established that the track is likely to be a target of interest.

The time a track is held in the preliminary stage is a programmable option and needs to be set to balance the speed of detection with the likelihood of a false alarm. In a low clutter environment, where extracted plots are likely derived from real targets, the acquisition time may be as short as 2 detections. For noisy situations, where the plot extractor is reporting false detections, the integration time in the preliminary stage may be extended.

Multi Hypothesis Tracking (MHT)

The SPx Tracker uses multi-hypothesis association. This offers significant improvements in perfomance over simpler single-hypothesis trackers. The role of a tracker is to interpret radar observations to distinguish real targets from noise, and to construct models to describe the motion of the true targets. The tracker is provided with data, typically in the form of plots, derived from the processing of the radar video. These plots are connected regions of radar video that satisfy some rules of position, amplitude, size and signal strength. Unfortunately, measurements from the radar are imperfect. There will be noise from the measurement process, clutter from the environment and unpredicted manoeuvres of the targets of interest. This means that the tracker will be presented with noisy and possibly multiple measurements from the target of interest. The tracker's responsibility is to provide the best interpretation of the data using assumed or calculated statistics for the noise and the likelihood of change.

In the single hypothesis situation, the tracker is forced to make the  interpretation it can of the available data at each update. For some updates, where there is a clear interpretation of the measurement, the best interpretation may be obvious and the single hypothesis offers a satisfactory solution. Problems arise, however, if the interpretation of the measurements is not obvious. In this case it may be desirable to defer a decision until the next update when additional information will help to decide on the correct interpretation. The ability to simultaneously consider multiple interpretations of the system is the key to the multi hypothesis tracker.

Track Filter

For each hypothesis, the tracker updates the current estimated position with the new measurement. If the measurement were known to be completely accurate, the update process would believe the measurement and the new estimate would be exactly the measured value. For various reasons, the measurement is inaccurate so the update process must take a weighted combination of the expected position and the measured position. This is the track filtering. SPx offers a number of track filtering modes. The simplest mode uses fixed gains in the components of the measurement. This can be successful for tracking applications where the target is clearly identified and relatively clutter free.

The filter works by computing a dynamic filter gain, K, based on estimated system noise and measurement noise models. The system noise is used to model uncertainty in the known dynamics of the target, including its ability to manoeuvre. As system noise increases, or equivalently as measurement noise decreases, the filter places more weight on the measurement so the filter gains increase. As system noise decreases or as measurement noise increases, the filter gains decreases causing less emphasis to be placed on the new measurement. The filter gains are continually changing and provide, under certain assumptions of the noise characteristics and linearity, an optimal estimation of the true target position.

Tracking Parameters

The behavior of the tracker may be configured through a set of tracking parameters. These parameters may be set initially from a configuration file and may be adjusted during operation of the tracker using either a customer GUI or network interface. The parameters control many aspects of the tracker's performance including:

· Min/max speed of target to be tracked

      · Multi or single hypothesis association modes

      · Fixed gain or adapative gain

      · Expected target dynamics

      · Size limitations on targets to be tracked

      · Measurement noise estimates