www.radartutorial.eu www.radartutorial.eu Radar Basics

Multiple Hypothesis Tracker (MHT)

Ambiguous association of radar plots with tracks 121 and 131
Track 121
Track 131

Figure 1: Ambiguous data association situation for the radar tracker

Ambiguous association of radar plots with tracks 121 and 131
Track 121
Track 131

Figure 1: Ambiguous data association situation for the radar tracker

Ambiguous association of radar plots with tracks 121 and 131
Track 121
Track 131

Figure 1: Ambiguous data association situation for the radar tracker

Multiple Hypothesis Tracker (MHT)

A Multiple Hypothesis Tracker (MHT) is an automatic radar target tracking method that considers several possible associations between new radar plots and existing tracks simultaneously. Unlike simple data association methods, the radar tracker does not have to commit immediately to a single association when the measurements are ambiguous. Instead, it can maintain several competing hypotheses about the targets' motion histories and continue evaluating them using subsequent radar measurements.

Such ambiguity may arise, for example, when two targets approach each other, their tracks cross, or several radar plots fall within possible association regions. In these situations, it may initially be unclear which plot belongs to which track. An immediate incorrect decision could cause the tracks to be swapped or the estimated motion histories to no longer correspond to the actual targets. The MHT attempts to avoid such errors by initially maintaining alternative associations in parallel.

Hypothesis Generation and Evaluation

With each new set of radar measurements, the possible associations of the detected plots with existing tracks are examined. A hypothesis describes a particular combination of these associations. It may also include assumptions that a target was temporarily not detected, a plot resulted from a false alarm, or a new track must be initiated. These different possibilities give rise to several alternative hypotheses, which the MHT initially maintains separately.

The hypotheses are evaluated according to how well they agree with the radar measurements and the predicted target motion states. This evaluation may consider, among other factors, the differences between measured and predicted positions, the uncertainties of the state estimates, and the probabilities of target detection and false alarms. Each hypothesis is therefore assigned a score that expresses its relative plausibility compared with the competing alternatives.

The key point is that a hypothesis does not have to be selected permanently after a single antenna rotation. Subsequent measurements may reveal that an association initially considered less likely is actually more consistent with the targets' true motion. The MHT therefore uses the sequence of measurements from several radar scans to make more reliable decisions about associations that were initially ambiguous.

Example of Ambiguous Data Association

In Figure 1, tracks 121 and 131 approach each other. If new radar plots are located in a region where either track is a possible association, the radar tracker may initially be unable to determine which measurement belongs to which target. One hypothesis associates the new plots according to the targets' previous motion histories. Another hypothesis assumes that the associations are reversed. The MHT can initially maintain both possibilities and compare their plausibility using subsequent measurements.

The resulting branches initially represent different assumptions about the association of measurements and the motion of the targets. They do not necessarily mean that an actual target has split into several separate targets. Similarly, maintaining several hypotheses simultaneously does not mean that a single radar plot may be assigned to multiple independent targets in the final track configuration. Instead, the different hypotheses represent competing possible explanations of the same measurement situation.

Limiting Computational Complexity

A major disadvantage of Multiple Hypothesis Tracking is its high demand for computing power and memory. Since each new radar measurement may introduce additional association possibilities, the number of possible hypotheses grows rapidly without additional limiting measures. Especially when many targets, numerous false alarms, or frequent missed detections are involved, evaluating every conceivable hypothesis may become computationally impractical.

For this reason, MHT systems employ methods to limit the number of hypotheses that are maintained. In Pruning, for example, hypotheses are discarded when their scores are substantially worse than those of the most likely alternatives. In Hypothesis Merging, sufficiently similar hypotheses may be combined under appropriate conditions. Further limitations can be imposed by specifying a maximum number of hypotheses or by restricting the depth of the measurement history being considered.

These measures reduce the computational burden, but they carry the risk of prematurely discarding a hypothesis that might later prove to be correct. The choice of limiting methods must therefore strike a balance between reliable target tracking and the available computational resources.

Track Output

Although the MHT internally maintains several alternative data association hypotheses, the radar system normally requires a single, unambiguous track configuration for further processing and display. For this purpose, the highest-scoring hypothesis may be selected. Depending on the implementation, other decision methods may also be used, such as deriving a track configuration from several evaluated hypotheses. Therefore, not all internally maintained alternatives need to appear on the radar display.

Multiple Hypothesis Tracking is particularly suitable for situations in which the association of individual radar measurements with existing tracks is uncertain. Its principal advantage is that ambiguous decisions do not have to be made immediately and permanently, but can instead be improved by evaluating subsequent measurements. This increases the likelihood of maintaining reliable automatic target tracking even when targets move close to one another or their paths cross.