The general problem is that current surveillance data is limited to real-time analysis by direct human monitoring or forensically through recording. The large amounts of data received from surveillance cameras pose a challenge for software to interpret potential threats beyond object and text recognition. A system is needed to automate monitoring processes and provide a predictive advantage to monitoring people by identifying potential threats from observed human activities. The specific technical problems we are addressing involve the need for technology to automate understanding of human behavior using contextual information, obtained by observing interactions in space and time. Image and video exploitation algorithms need to first decompose incoming visual information, distinguishing humans from natural and man-made structures. Further computation must then discover human activity based on motion, posture and body positioning. Potential threats are identified based on predetermined hostile human activity. The resulting classifications of detected threat events are visually communicated to the end user. VIPMobile is developing a lean and robust software package that exploits incoming video data to interpret human behavior with systematic analysis of spacial-temporal block algorithms, motion detection, tracking and outlier identification. The system also utilizes a database of predefined hostile behavior factors that processes feature extraction and edge detection to visually communicate anomalous activity to monitoring personnel in near real time so preventative actions can be taken.
Keywords: Image Processing, Behavior, Data Fusion, Real-Time, Cognitive Science
