Intelligent Video Surveillance and Analysis

UtopiaCompression develops innovative, cutting edge technologies for Intelligent Video Surveillance (IVS) systems. UC’s approach to IVS is based on a suite of innovative Visual Knowledge Discovery (VKD) technologies that are capable of mining highly semantic information from colossal volumes of surveillance video/image data for automated scene understanding and threat detection and identification.  The technology suite includes the three main components:

1. The Image/Video parser that provides advanced performance for automatic target detection, tracking and classification in real-time video streams, and can work in a broad range of surveillance environments.
2. Activity recognition: The various visual cues and features obtained from the image/video parser can be fused together with other types of data/information/knowledge available to achieve fully automated visual activities recognition and subsequently assist the complex task planning and decision making processes. 
3. Content-based indexing and retrieval tools that enable rapid semantic search of captured video data and to perform forensic analysis.

Detection of targets in maritime scenes

Target Detection

Based on advanced image processing technologies, UC has developed a suite of highly accurate and efficient target detectors capable of detecting targets of interests for a variety of surveillance conditions of background environments (maritime, aerial, urban etc), sensor types (EO and IR), and sensor condition (stationary and moving).

Multi-Target Tracking Technology

UC has developed a real-time robust object tracking system using EO/IR cameras. The UC method can track the identity of multiple targets stably for long time and robustly against occlusions, clutter, lighting changes, crowded scenes, and varying number of targets.

Example of handling target occlusion

Hierarchical Target Recognition and Classification

UC target classification technology is capable of recognizing a target and determining its class within a large hierarchical ontology of object concepts. UC technology offers significant advantages over other state-of-the-art target recognition methods in terms of high accuracy, very fast classification time, and the large number of classes that can be recognized.

Activity Recognition

For achieving intelligent video analysis, UC’s VKD technologies support robust recognition of complex visual activities/events by parsing various visual features/cues and fusing them with the proper context information and knowledge database.  The activity recognition system capitalizes on efficient representation of domain-specific knowledge and allows robust concurrent event analysis (even for scenarios with uncertainties and errors).  Recognized potential threats will trigger automatic generation of alerts and messages to notify the personnel in-the-loop for proper attentions/actions.

Intelligent Video Retrieval

Surveillance imagery data can be efficiently searched and retrieved using UC semantic-based video indexing and retrieval technology. The system allows user to rapidly search large corpus of surveillance videos for intelligence information. The user can search the videos either by using semantic concepts of target/events or supplying a visual example. UC video indexing and retrieval technology saves enormous time of forensic analyst watching and manually tagging the video collected, which could otherwise lead to errors and inconsistence.

Image retrieval results for semantic concept “Sloop”

Image retrieval by example (QBE)

The capabilities of UC’s VKD technologies are demonstrated for maritime surveillance systems in the following government-funded programs:

1. Development of a video security shell for Littoral Combat Ship (LCS): In this application, the environment surrounding a Littoral Combat Ship (LCS) is constantly monitored and analyzed.  Objects of interests (i.e. boats and humans in this case) are automatically detected, tracked and classified.  Symbols representing various visual cues at different levels are generated and analyzed.  Any suspicious event/activity will be automatically recognized and the corresponding alert will be sent, as demonstrated in the video clip.
2. Intelligent Retrieval of Surveillance Imagery: Develop a semantic based video search engine that enable users of video surveillance systems to easily conduct video-based forensics and Imagery Intelligence (IMINT) from video imagery captured in harbor and sea shores.