What Is Tracking?

Tracking is a field of signal processing and control concerned with estimating and following the state of a moving or changing object over time from a sequence of measurements. It draws on probability theory, estimation theory, and control engineering to form a continuous model of position, velocity, orientation, or other dynamic quantities from sensor data that is incomplete, noisy, and sometimes missing entirely. Applications range from radar and sonar surveillance to computer vision, robotics, and biomedical imaging.

The core challenge of tracking is that no sensor directly reveals the true state of a target. Each measurement is a probabilistic observation, corrupted by receiver noise and the environment. Tracking algorithms maintain a probabilistic belief about where the target is and how it is moving, updating that belief recursively as new observations arrive. Maximum likelihood estimation and Bayesian inference are foundational mathematical frameworks that underlie most practical tracking systems.

State Estimation and Filtering

The Kalman filter is the standard algorithm for tracking targets whose motion obeys a linear model and whose measurement noise is Gaussian. Introduced by Rudolf E. Kalman in 1960, the filter alternates between a prediction step, which projects the current state estimate forward using a motion model, and an update step, which corrects the prediction using the incoming measurement. For nonlinear systems, extensions including the Extended Kalman Filter and the Unscented Kalman Filter linearize or approximate the nonlinear dynamics, while still retaining the recursive Gaussian structure. Research published in IEEE Xplore on Kalman filter-based target tracking demonstrates how these algorithms scale to multi-target scenarios where the number of objects being tracked is itself unknown and changing.

Multiple-hypothesis tracking (MHT) and joint probabilistic data association (JPDA) extend the filtering framework to situations where measurement-to-target assignment is ambiguous, handling crossing trajectories and clutter returns. These algorithms are central to air traffic control and military surveillance, where false alarms from background returns must be rejected without discarding genuine tracks.

Particle Tracking

Particle tracking replaces the Gaussian approximation of the Kalman filter with a Monte Carlo representation of the probability distribution over target states. A large set of weighted samples, called particles, each represents a hypothesis about the current state. After each measurement, particles are reweighted according to how well they match the observation, and low-weight particles are replaced by resampling around high-weight ones. The particle filter handles arbitrary nonlinear motion models and non-Gaussian noise distributions, making it well suited to visual tracking under illumination change, occlusion, and cluttered backgrounds. In biology and medical imaging, particle tracking algorithms are used to follow individual molecules or cells across fluorescence microscopy frames, yielding quantitative measurements of diffusion and transport.

Learning-Based and Iterative Tracking Methods

Iterative learning control (ILC) provides an alternative framework for tracking applications where the same trajectory is repeated multiple times. Rather than estimating a hidden state, ILC treats each trial as a learning iteration, updating the control signal by analyzing the error from the previous trial. Over successive repetitions, the tracking error converges toward zero even in the presence of model uncertainty, provided the trajectory is periodic. This approach is used in precision manufacturing, robotic assembly, and hard-disk read/write head positioning. Separately, deep learning has become prominent in video and radar tracking, with convolutional and recurrent networks trained to produce target-state estimates directly from raw sensor data. The NIST Interagency Report on object detection and tracking reviews evaluation methodologies used to benchmark tracking performance across these different algorithmic families. Survey work on deep visual object tracking covers the evolution of learning-based methods from correlation filters through transformer architectures.

Applications

Tracking has applications in a wide range of fields, including:

• Radar and sonar surveillance for air and maritime traffic management
• Autonomous vehicle perception and collision avoidance
• Biomedical cell and molecular tracking in fluorescence microscopy
• Robotics and mobile platform navigation
• Sports analytics and broadcast motion capture