Scientists use AI to convert images of people into real-time avatars
n n
Through artificial intelligence, a team of scientists has developed software to create avatars of crowds of people filmed with a video camera, giving commercial facilities and transportation operators information on congestion and other data in real time.
rn
The technology can identify the clothing colors and motions of shoppers and passers-by to make their graphical representations, and display how many people are in locations as well as what kind of individuals are present. Through the use of avatars, this can be done without invading people’s privacy.
rn
Working with the Tokyo-based Japan Image Analysis Association, an electronics maker and other parties, Hironori Yamauchi, a professor of image recognition engineering at Ritsumeikan University’s College of Science and Engineering, invented the software.
rn
“If security cameras are equipped with the software, the congestion degree and atmosphere of stores can be shown in public places without invading people’s privacy,” Yamauchi said. “It is therefore expected to help ease congestion at commercial facilities and roads.”
rn
Under the system, raw images taken by cameras are examined by AI to detect people so that all individuals in the images can be converted into computerized alter egos.
rn
Not only the colors of people’s hair and clothing but also their postures and ways of walking are categorized into groups, enabling those features to be reproduced in the avatars.
rn
Hiroshi Miyanaga, a technology management professor at the graduate school of the Tokyo University of Science, who is well-versed in image processing and commercial use of images, called the invention “the pioneering software among other commercially applicable technologies.”
rn
The research team said it will develop an improved version of the software to reflect filmed individuals’ height, body size and gender in their representations by analyzing such personal features with AI by the summer at the earliest.
rn
Under the system, only characteristics whose collection would not violate the personal information protection law will be displayed so that filmed individuals cannot be identified.
rn
“Introducing multiple cameras to record only avatars will make it possible to display slight changes in the number of passers-by in numerical format in real time,” Miyanaga said. “For example, places that are popular among women can be more promptly spotted and the popularity of shops evaluated quantitatively.
rn
“The technology will thus bring significant value to commercial facilities and real estate agents, as they can currently rely only on employees’ gut instinct and questionnaire surveys.”
rn
Miyanaga also said the new system can be used for a wide range of purposes such as attending online meetings and receiving online counseling services anonymously.
rn
Conventional video camera footage cannot precisely express the depth of space, so existing image analysis systems often fail to track individuals when there are many passers-by around.
rn
The recently developed system can identify and recognize each individual, making it possible to spot a person who leaves a suspicious object behind in crowded areas though the task is difficult for conventional technology, according to the scientists.
rn
The researchers said the new technology can be used to help identify who the avatars represent by closely analyzing limb and body movements of walking people. As individuals can be identified and recognized at lower costs under the method, the research team is also developing a security system separately from the software.
rn
“Cheaper and more precise surveillance mechanisms can be established (based on the new technology),” said Miyanaga. “Because of that, public discussions may be needed over raw image deletion and details of data to be reflected in some fields of application.”
n n