Seeing and Understanding Images Like the Human Eye

Deep View, an Engine That Understands Behaviors

It is not a difficult task for humans to distinguish dogs from cats, recognize a human face, or find out what a person is doing. The field of image recognition, which corresponds to the human "visual intelligence," is developing through deep learning. Since the development of Google's artificial intelligence, which detected cats in YouTube videos in 2012, and the artificial intelligence developed by University of Toronto, which could recognize objects in images with a greatly improved accuracy, innovative developments are continuously being made.

01

Deep View,
recognizing and analyzing data
in images like humans

What would it be like if a machine could recognize a scene and make judgments like a human? ETRI’s SW Content Research Center is working on the answer to that question. The ETRI SW Content Research Center studies and develops a high-performance visual discovery platform (Deep View) that understands and predicts large-scale image data in real-time. It is a technology that constitutes an "image Big Data platform" that can recognize people and objects in images, and understand the meaning of them as accurately as humans can.

The technology helps us understand changes that occur in urban spaces, and predict disasters and crimes in urban spaces in real time through large-scale collection of images and videos. In addition, it constructs the large-scale visual Big Data that we need, enabling stable information analysis at the national level.

Conventional academic research on understanding human behaviors experienced a limit in recognizing actual human behaviors as it was developed from general data such as sports videos and YouTube videos. In order to solve this issue, ETRI improved necessary functions, requirements, and data through cooperation with local governments, and focused on developing a behavior-understanding technology that can actually be implemented in real environments. .

02

Implementing
the technology in real life

As the first step of commercialization of Deep View, ETRI used CCTV footage, utilizing Deep View to monitor illegal dumping in urban areas. CCTVs equipped with Deep View were installed in some areas of Sejong city, and downtown Eunpyeong-gu, Seoul, to monitor illegal dumping. Unlike the conventional CCTVs that simply recognize presence of people, the CCTVs equipped with Deep View recognized particular actions of people, such as throwing or putting objects down, and sent out a warning message to prevent illegal dumping.

The CCTVs equipped with Deep View detect human actions by detecting the locations of human joints and objects, and then drawing a model that reflects the relationship between the person and the object. In addition, it detects a garbage heap and recognizes the human act of dumping garbage. Based on this information, it traces and infers a relationship as well. It also detects various patterns of behavior related to waste dumping, whether being a certain distance away, throwing garbage, or discharging garbage completely.

Monitoring illegal dumping is just the beginning. Once the Deep View technology is commercialized, it is expected to secure fundamental technologies that will advance the national social safety net. It could be also used to improve public safety, national security, and convenience in various areas such as protecting citizens in urban areas through CCTVs. ICT technology that has visual intelligence will keep our cities safe and comfortable, ultimately increasing quality of our life.

03

ETRI increases possibilities of AI

ETRI was selected as the operator of the visual AI platform technology development project initiated by the Ministry of Science and ICT until 2024. ETRI also plans to develop technologies specialized for public services such as the urban crime prevention system of the Ministry of the Interior and Safety and highway CCTVs.

Park Jong-yeol, Head of the Visual Intelligence Research Group, who is in charge of this technology, commented, “Deep View is a technology that is being actively developed by leading global companies as well. The technological prowess of the ETRI research team has already received global recognition by winning the second place at the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) last year in the Object Detection (DET) category. We will further develop services that can be applied to the relevant technologies in the public sector." In the future, the SW Content Research Institute will continue to act as an eye that monitors safety of all citizens by securing the analysis function for images and videos that are generated in the national infrastructures in order to secure social safety, which is a national requirement.

With the long-term acquisition of artificial intelligence technology that can see, hear and learn on its own like humans as its goal, the Intelligence Information Research Division specializes in language intelligence, voice intelligence, visual intelligence, and smart data fields, which are key drivers of the age of the Fourth Industrial Revolution. It is also studying original technologies related to complex intelligence and the ever-growing artificial intelligence in order to prepare for the next-generation technology after Deep Learning. The Intelligence Information Research Division intends to secure key intelligence information technologies through selection and focus, and support the ecosystem by supplying and spreading the acquired technologies through an open API platform in order to revitalize the Korean intelligence information industry.

List of 45 Key Research Accomplishments