SeaAlert: Video Analysis of Drug and Human Smuggling Activities

Sky Park Labs proposes SeaAlert, a video event detection tool that automatically detects anomalous events and identifies specific activities of interest. SeaAlert learns spatial and temporal patterns in a video to establish common objects and behaviors in a scene. The system detects and tracks objects of interest in the scene and analyzes the objects to extract a set of characteristic features that describe each target’s appearance, and dynamics. A statistical model of normal activity of the scene is learned by analyzing the density of these features that constitute the most prevalent object types, locations, and trajectories for object categories of interest. In addition, these features can be used to detect specific events, such as people jumping from a vessel, using sequence models. Once trained, the software processes incoming video and proactively identifies anomalous or specifics events using the trained models. An interface allows the analyst to visualize and adjudicate detected events quickly. We hypothesize SeaAlert can model complex events involving the interaction of multiple object types, such as a people departing a vessel that landed on shore, using the rich event feature representation. The object detection algorithm achieves high detection accuracy across various imaging modalities (IR and color). Also, the algorithms are computationally efficient, capable of processing video at 30 FPS on resource constrained devices, making it amenable to deployment in a variety of maritime operational environments. SeaAlert represents a substantial breakthrough in video analytics, providing benefits to a wide array of applications including retail, healthcare, and sports analytics.