Technology has been introduced that allows videos captured from CCTV, etc. to be used. for learning using artificial intelligence, in compliance with the Personal Information Protection Act.
[디지털투데이 박건도 기자]Dipping Source has actively taken measures to expand the technology that can be used for AI analysis of videos captured by CCTV, etc., in compliance with the Personal Information Protection Law.
According to the current data protection law, original videos cannot be used for artificial intelligence learning. Therefore, most companies use scripted videos for AI learning. However, the cost of producing videos is high, making it difficult to learn video AI at scale.
Dipping Source points out that its SEAL solution can solve this problem.
Dipping Source introduced the thread in detail in a conference hosted by the Ministry of Science and ICT and the Korea Data Agency on the 14th.
Kim Tae-hoon, CEO of Dipping Source, said in his presentation “Innovation in offline retail through video analytics on personal information protection”, “Artificial intelligence learning is possible without violating current laws by removing information personally identifiable information (PII) such as iris, filter and fingerprint of the original video.” “The SEALs intentionally modify the information contained in the video data into obfuscated data to protect personal information,” he said.
SEAL appears to simply mosaic personal information into the video data, but contrary to what it seems, Dipping Source explained that it actually retains enough information needed for the AI to learn. This is in contrast to simple mosaic processing of video, which has little use of artificial intelligence.
As a result of learning a pedestrian detection model with a public dataset of 15,000 images, Dipping Source reported that if the mean accuracy (mAP) was 0.57 for the original image, SEAL was 0.569 and the method of the simple mosaic it was 0.45.
A Dipping Source official said, “If deidentification is performed using the existing mosaic method, data collection and processing are possible, but AI model learning and inference is difficult.” He added, “SEAL produces almost the same AI learning results as the original video.” He said.
This technology is used in offline retail stores in Korea, the United States and Japan. CEO Kim Tae-hoon said the products were supplied to COEX and global retail companies.
SEAL is capable of anonymization in real time. The immersion source AI vision sensor connected to CCTV detects subjects in the video and measures interest, movement, etc. Dipping Source explained that the information is analyzed and provided to customers in the form of dashboards for purposes such as marketing and crime prevention.
CEO Kim said: “Using CCTV cameras already installed in stores, we accumulate data about people or objects in real time to provide the information needed for business decision making,” adding: “Personal identification is absolutely impossible, which is has always been a controversial issue in AI video analytics.” “We solved the problem with the SEALs,” he said.
#Kim #Taehoon #CEO #Dipping #Source #CCTV #footage #analyzed #artificial #intelligence #violating #personal #information
