Video Surveillance and Tracking Methods

Video surveillance and tracking methods address how computers can reliably follow objects or people across video frames, often through crowded scenes, changing lighting, or partial occlusions. Core techniques range from classical background subtraction, which separates moving foreground elements from a static scene, to deep learning approaches using convolutional neural networks that learn rich visual representations for person re-identification across non-overlapping cameras. The practical stakes are high — applications span traffic monitoring, public safety, and autonomous navigation — but so are the challenges: systems must operate in real time while handling dramatic appearance changes, and they must generalize to environments and identities they were never trained on. Active research directions include making multi-object tracking robust without dense manual annotation and building methods that are accurate yet transparent enough for accountable deployment.

Works

82,985

Total citations

1,449,061

Keywords

Visual TrackingObject TrackingPerson Re-identificationBackground SubtractionConvolutional Neural NetworksReal-time Tracking