Probabilistic Vehicular Trace Reconstruction Based on RF-Visual Data Fusion

Geolocation information is not only crucial in conventional crime investigation, but also increasingly important for digital forensics as it allows for the logical fusion of digital evidence that is often fragmented across disparate mobile assets. This, in turn, often requires the reconstruction of mobility patterns. However, real-time surveillance is often difficult and costly to conduct, especially in criminal scenarios where such process needs to take place clandestinely. In this paper, we consider a vehicular tracking scenario and we propose an offline post hoc vehicular trace reconstruction mechanism that can accurately reconstruct vehicular mobility traces of a target entity by fusing the corresponding available visual and radio-frequency surveillance data. The algorithm provides a probabilistic treatment to the problem of incomplete data by means of Bayesian inference. In particular, we realize that it is very likely that a reconstructed route of a target entity will contain gaps (due to missing trace data), so we try to probabilistically fill these gaps. This allows law enforcement agents to conduct off-line tracking while characterizing the quality of available evidence.