The Action Recognition Group researches methods to determine the "where" and "what" of people in an image, focusing on three main technologies
Behavior analysis aims to preempt crimes and quicken response times to incidents by detecting criminal and abnormal behavior.
Recently, great strides in research in crowd behavior analysis have been made due to an increased emphasis on wide-area security for large venues. At the 2017 Tokyo Marathon, research results including crowd density estimation and crowd movement analysis were trialed, leading to a reduction in the workload of security personnel.
Since the late 90's, SECOM has continually released products incorporating video-based processing, such as the image-based intrusion sensor "SECOM AX," the behavior-based "Intelligent Emergency Alert System," and the "Walk-through Face Recognition System." All of these products are world-firsts in their respective categories in incorporating video, and all of these systems would not have been possible without including technologies developed at Intelligent Systems Laboratory. Furthermore, these technologies include advanced pattern recognition algorithms, which are the core of modern artificial intelligence (AI).
Intelligent Systems Laboratory has been involved in commercializing AI since before the late 90s which is regarded as the advent of deep learning. By merging recent advances in deep learning with our knowledge and technologies, we believe that we can create innovative new SECOM services.
Behavior analysis requires the fusion of multiple technologies. When monitoring outside, cars and other objects must in analyzed, in addition to humans. When monitoring people, estimating attributes such as age and gender provide better insight into behavior.
Behavior analysis requires coordinating between multiple groups within the Vision Intelligence Division. For example, algorithms developed by the Target Detection Group detect and track people and vehicles. The facial attributes of the detected people can be determined through the work of the Attribute Analysis Group to uniquely identify people, their gender and approximate age.
All of this research is supported by the fundamental research into camera calibration and environmental modeling conducted by the Vision Processing Group. Finally, the Fusion Architecture Group conducts research into methods to commercialize these technologies, focusing on system reliability and stability.