Target tracking is conventionally performed using point measurement (PM) outputs from a sensor signal detector. However, when tracking targets with low signal to noise ratios (SNR), a low threshold for detection is set to minimise information loss.

This setting increases in the number of false alarms generated, and accurate track-to-measurement data association becomes difficult. Although superior tracking of low SNR targets can be achieved by employing track-before-detect algorithms, they are computationally intensive.

In this paper, a computationally efficient multi-target tracking algorithm is proposed by exploiting signal strength within a conventional PM based Bayesian framework for tracking low SNR targets. The proposed filter employs a δ-generalised labelled multi-Bernoulli filter with signal strength and PMs. By employing a Rao-Blackwell decomposition, the filtering problem is efficiently rendered in two parts: (i) a particle filter to estimate the signal strength, and (ii) an analytical filter to estimate the target motion state. Performance is analysed for a case study where PMs are bearing and range, signal strength clutter is modelled by a Gamma distribution, and the target signal return is a narrowband Swerling II model. We compare the proposed filter against a δ-generalised labelled multi-Bernoulli filter with conventional point measurements only.

Numerical results demonstrate that signal strength information in the proposed filter facilitates accurate identification of target detections from measurement sets densely populated with false alarms. This improves the ability to establish, maintain, and eliminate target tracks as required when compared to the equivalent δ-GLMB filter without signal strength information. Under very low SNR conditions, the additional signal strength information enables consistent identification and tracking of targets that are otherwise repeatedly missed. Computationally, the proposed filter is approximately twice as demanding as the equivalent filter without signal strength information, which is below expectations. This is attributed to the additional information enabling more efficient management of multi-target hypotheses.

This research proposes a novel technique that has the potential to enhance the performance of sensor tracking systems in scenarios with low SNR measurements. These conditions are prevalent in real-world applications, making the proposed technique a valuable contribution to the field. By addressing the challenges posed by low SNR environments, this research improves tracking accuracy and reliability in various application areas that rely on sensor tracking systems.

Key Takeaways:

• Enhanced tracking performance for low signal-to-noise ratio targets in the presence of numerous false alarms.
• Signal strength information is shown to contain key information to robustly differentiate between target measurements and noise-induced false alarms.
• Tracking framework is suitable for deployment on resource-constrained computing platforms.
• Integrity of downstream systems is improved by increased tracking reliability and accuracy.