Abstract:
Track-to-track association has emerged as a signi cant problem to extract critical information
from the tracks shared by multiple sensors interconnected through a wireless network.
e information shared in the form of packets over a network contains biases that results in
deviation of the object position from its real position. e tracks obtained are initially preprocessed
to analyze, compute and update the state co-variance matrix of each sensor in Earth
Centered Earth Fixed frame, which is a function of measurement errors, process model(s) and
the Markov chain transition matrix. In inclusion, a track synchronization lter is used to deal
with the delays. A linear lter is used to predict the objects state at a speci c time before proceeding
to the track-to-track association stage. An unsupervised machine learning technique
is proposed for track-to-track association problem in a network of multiple sensors and objects
in the presence of sensor noises. e proposed association technique requires state vector and
the corresponding state co-variance matrix. e approach involves classifying the objects data
in the form of clusters using a variant of hierarchical agglomerative clustering algorithm. is
classi cation is o en based on a prede ned threshold which is selected from past data of similar
scenarios. A more realistic approach is proposed for threshold calculation that updates
with the current object data and is independent of past scenarios data. e proposed threshold
estimation technique is put to the test against a variety of scenarios, ranging from simple to
extremely complex. Furthermore, the accuracy of the proposed algorithm is determined by
implementing it against multiple complex and realistic environments. e e ect of alternating
systematic and random biases on the accuracy of the proposed algorithm is analyzed against
range, azimuth, and elevation of object.
