It is becoming increasing clear that it is humanly impossible to analyze a deluge of data from cameras and other sensors in a variety of applications including surveillance, railroad inspection, driver as sistance. The practical systems that we built, although in pursuit of different business objectives, share a common goal, which is to intelligenthy and efficiently analyze and extract the most important actionable information from an overwhelming amount of data, while being able to effectively ignore a large portions of uneventful and/or noisy data. This presentation will focus on computer vision, machine learning and system optimization techniques that we used to successfully address different technical and business challeges, and to deliver differentiating performance to meet our customers expectations