Adaptive, Automated Real-time Event/Target Detection using Supervised Learning

Human operators must closely monitor video for simultaneous situational awareness and threat assessment. For instance, urban environments in a state of constant activity generate numerous visual cues, each of which must be examined so that potential security breaches do not go unnoticed. The need for constant vigilance places a significant burden on the human operator, invariably leading to fatigue and lapses in attention span. Vecna proposes a video event detection software tool, known as AESOP, that automatically detects time critical events in real-time. AESOP learns new events using a programming by example. With this technique, the analyst teaches the software tool new events by demonstrating actions on concrete examples. Once trained, AESOP processes incoming video and proactively identifies user-defined events in real-time while also indexing the video to simplify forensic analysis. Using state-of-the-art computer vision algorithms, the system identifies and tracks all targets in the scene. Characteristic features for each target are extracted over time yielding feature trajectories which are then efficiently matched to the trained event trajectories. An easy-to-use user interface allows the analyst to visualize and confirm detected events quickly. Furthermore, AESOP's accuracy can improve by incorporating the confirmed detected events into the training set.