Kaist

Why Don’t My Document Photos Rotate Correctly？

  〈 The team of Professor Lee and his Ph.D.student Jeungmin Oh  developed a technique that can correct a phone’s orientation by tracking the rotation sensor in a phone.) 〉      John, an insurance planner, took several photos of a competitors’ new brochures. At a meeting, he opened a photo gallery to discuss the documents with his colleagues. He found, however, that the photos of the document had the wrong orientation; they had been rotated in 90 degrees clockwise. He then rotated his phone 90 degrees counterclockwise, but the document photos also rotated at the same time. After trying this several times, he realized that it was impossible to display the document photos correctly on his phone. Instead, he had to set his phone down on a table and move his chair to show the photos in the correct orientation. It was very frustrating for John and his colleagues, because the document photos had different patterns of orientation errors.     Professor Uichin Lee and his team at KAIST have identified the key reasons for such orientation errors and proposed novel techniques to solve this problem efficiently. Interestingly, it was due to a software glitch in screen rotation?tracking algorithms, and all smartphones on the market suffer from this error. When taking a photo of a document, your smartphone generally becomes parallel to the flat surface, as shown in the figure above (right). Professor Lee said, “Your phone fails to track the orientation if you make any rotation changes at that moment.” This is because software engineers designed the rotation tracking software in conventional smartphones with the following assumption: people hold their phones vertically either in portrait or landscape orientations. Orientation tracking can be done by simply measuring the gravity direction using an acceleration sensor in the phone (for example, whether gravity falls into the portrait or landscape direction). Professor Lee’s team conducted a controlled experiment to discover how often orientation errors happen in document-capturing tasks. Surprisingly, their results showed that landscape document photos had error rates of 93％. Smartphones’ camera apps display the current orientation using a camera-shaped icon, but users are unaware of this feature, nor do they notice its state when they take document photos. This is why we often encounter rotation errors in our daily lives, with no idea of why the errors are occurring.     The team developed a technique that can correct a phone’s orientation by tracking the rotation sensor in a phone. When people take document photos their smartphones become parallel to the documents on a flat surface. This intention of photographing documents can be easily recognizable because gravity falls onto the phone’s surface. The current orientation can be tracked by monitoring the occurrence of significant rotation.       In addition, the research team discovered that when taking a document photo, the user tends to tilt the phone, just slightly, towards the user (called a “micro-tilt phenomenon”). While the tilting degree is very small?almost indistinguishable to the naked eye?these distinct behavioral cues are enough to train machine-learning models that can easily learn the patterns of gravity distributions across the phone. The team’s experimental results showed that their algorithms can accurately track phone orientation in document-capturing tasks at 93％ accuracy. Their approaches can be readily integrated into both Google Android and Apple iPhones. The key benefits of their proposals are that the correction software works only when the intent of photographing documents is detected, and that it can seamlessly work with existing orientation tracking methods without conflict. The research team even suggested a novel user interface for photographing documents. Just like with photocopiers, the capture interface overlays a document shape onto a viewfinder so that the user can easily double-check possible orientation errors. Professor Lee said, “Photographing documents is part of our daily activities, but orientation errors are so prevalent that many users have difficulties in viewing their documents on their phones without even knowing why such errors happen.” He added, “We can easily detect users’ intentions to photograph a document and automatically correct orientation changes. Our techniques not only eliminate any inconvenience with orientation errors, but also enable a range of novel applications specifically designed for document capturing.” This work, supported by the Korean Government (MSIP), was published online in the International Journal of Human-Computer Studies in March 2017. In addition, their US patent application was granted in March 2017.

Winning Best in Theme Award in NASA RASC-AL

〈 KAIST team of the Department of Aerospace Engineering poses after winning the Best in Theme Award in NASA's RASC-AL) 〉 〈 Prof. Jaemyung Ahn 〉   A students team from the Department of Aerospace Engineering won the Best in  Theme Award for moon exploration system design at Revolutionary Aerospace Systems Concepts - Academic Linkage (RASC-AL), an aerospace mission system design competition organized by NASA in the USA. The KAIST team, consisting of Jaeyoul Ko, Jongeun Suh, Juseong Lee, Sukmin Choi, and Eunkwang Lee, and supervised by Professor Jaemyung Ahn, competed as a joint team with Texas Tech University and the Royal Melbourne Institute of Technology in Australia, The joint team was selected as one of the 14 finalists after two preliminary rounds. The finals of RASC-AL Forum took place from May 30 to June 3 in Florida. The team received the top prize with their design entitled ‘Earth to Lunar Interchangeable Transportation Environment (ELITE) for Logistics Delivery Systems’, one of the four themes of the competition. Since 2002, RASC-AL competitions, managed by NASA, have been held with themes on innovative aerospace system and missions, in which world-class undergraduate and graduate students have participated. This year’s themes were ▲ Lightweight Exercise Suite ▲ Airlock Design ▲ Commercially Enabled LEO/Mars Habitable Module and ▲ Logistics Delivery System. Moon exploration requires a great deal of time and supplies. The KAIST team has been researching supply delivery systems in space for long-term manned moon exploration with their joint team for the last eight months. In particular, incidents can occur during the initial stages of long-term manned moon exploration missions that are unpredictable during system design and planning. Therefore, to cope with such unpredictability in the mission, the KAIST team deduced a system and an operational concept with increased flexibility to maximize the cost effectiveness of the supply transport. The spacecraft was divided into propulsion and transport modules based on their functionalities, and can allow the flexibility by switching the transport module according to the demands of the moon base. The operational flexibility and cost effectiveness are further increased by introducing multiple departure orbits from the Earth (e.g. low Earth orbit vs. geosynchronous Earth orbit) enabled by utilization of various launch vehicles. Professor Ahn, the advisor for the team, said, “I am proud of the students who collaborated with the international joint teams and achieved great result.” He continued, “I believe this to be the result of continuous efforts and initiatives of the department for system design-centered education. We will keep providing high-quality system design and education through various opportunities such as international cooperation in design education.”

Face Recognition System “K-Eye” Presented by KAIST

    Artificial intelligence (AI) is one of the key emerging technologies. Global IT companies are competitively launching the newest technologies and competition is heating up more than ever. However, most AI technologies focus on software and their operating speeds are low, making them a poor fit for mobile devices. Therefore, many big companies are investing to develop semiconductor chips for running AI programs with low power requirements but at high speeds. A research team led by Professor Hoi-Jun Yoo of the Department of Electrical Engineering has developed a semiconductor chip, CNNP (CNN Processor), that runs AI algorithms with ultra-low power, and K-Eye, a face recognition system using CNNP. The system was made in collaboration with a start-up company, UX Factory Co. The K-Eye series consists of two types: a wearable type and a dongle type. The wearable type device can be used with a smartphone via Bluetooth, and it can operate for more than 24 hours with its internal battery. Users hanging K-Eye around their necks can conveniently check information about people by using their smartphone or smart watch, which connects K-Eye and allows users to access a database via their smart devices. A smartphone with K-EyeQ, the dongle type device, can recognize and share information about users at any time. When recognizing that an authorized user is looking at its screen, the smartphone automatically turns on without a passcode, fingerprint, or iris authentication. Since it can distinguish whether an input face is coming from a saved photograph versus a real person, the smartphone cannot be tricked by the user’s photograph. The K-Eye series carries other distinct features. It can detect a face at first and then recognize it, and it is possible to maintain “Always-on” status with low power consumption of less than 1mW. To accomplish this, the research team proposed two key technologies: an image sensor with “Always-on” face detection and the CNNP face recognition chip. The first key technology, the “Always-on” image sensor, can determine if there is a face in its camera range. Then, it can capture frames and set the device to operate only when a face exists, reducing the standby power significantly. The face detection sensor combines analog and digital processing to reduce power consumption. With this approach, the analog processor, combined with the CMOS Image Sensor array, distinguishes the background area from the area likely to include a face, and the digital processor then detects the face only in the selected area. Hence, it becomes effective in terms of frame capture, face detection processing, and memory usage. The second key technology, CNNP, achieved incredibly low power consumption by optimizing a convolutional neural network (CNN) in the areas of circuitry, architecture, and algorithms. First, the on-chip memory integrated in CNNP is specially designed to enable data to be read in a vertical direction as well as in a horizontal direction. Second, it has immense computational power with 1024 multipliers and accumulators operating in parallel and is capable of directly transferring the temporal results to each other without accessing to the external memory or on-chip communication network. Third, convolution calculations with a two-dimensional filter in the CNN algorithm are approximated into two sequential calculations of one-dimensional filters to achieve higher speeds and lower power consumption. With these new technologies, CNNP achieved 97％ high accuracy but consumed only 1/5000 power of the GPU. Face recognition can be performed with only 0.62mW of power consumption, and the chip can show higher performance than the GPU by using more power. These chips were developed by Kyeongryeol Bong, a Ph. D. student under Professor Yoo and presented at the International Solid-State Circuit Conference (ISSCC) held in San Francisco in February. CNNP, which has the lowest reported power consumption in the world, has achieved a great deal of attention and has led to the development of the present K-Eye series for face recognition. Professor Yoo said “AI - processors will lead the era of the Fourth Industrial Revolution. With the development of this AI chip, we expect Korea to take the lead in global AI technology.” The research team and UX Factory Co. are preparing to commercialize the K-Eye series by the end of this year. According to a market researcher IDC, the market scale of the AI industry will grow from ＄127 billion last year to ＄165 billion in this year.   (Photo caption: Schematic diagram of K-Eye system)

KAIST Team Wins Bronze Medal at Int＇l Programming ..

  〈 Professor Taisook Han and his students 〉     A KAIST Team consisting of undergraduate students from the School of Computing and Department of Mathematical Science received a bronze medal and First Problem Solver award at an international undergraduate programming competition, The Association for Computing Machinery-International Collegiate Programming Contest (ACM-ICPC) World Finals. The 41st ACM-ICPC hosted by ACM and funded by IBM was held in South Dakota in the US on May 25. The competition, first held in 1977, is aimed at undergraduate students from around the world. A total of 50,000 students from 2900 universities and 103 countries participated in the regional competition and 400 students competed in the finals. The competition required teams of three to solve 12 problems. The KAIST team was coached by Emeritus Professor Sung-Yong Shin and Professor Taisook Han. The student contestants were Jihoon Ko and Hanpil Kang from the School of Computing and Jongwoon Lee from the Department of Mathematical Science. The team finished ranked 9th, receiving a bronze medal and a ＄3000 prize. Additionally, the team was the first to solve all the problems and received the First Problem Solver award. Detailed score information can be found on. https://icpc.baylor.edu/scoreboard/