A technology has been developed that may easily detect a hit-and-run vehicle that crashes a automobile parked in a parking zone and runs away using artificial intelligence (AI) technology.

Gwangju Institute of Science and Technology (President Lim Ki-cheol) announced on the 18th that Professor Lee Yong-gu's research team within the Department of Mechanical Engineering succeeded in detecting the time of a hit-and-run while driving from all CCTV images through artificial intelligence (AI) technology.

The research team developed a technology to detect the time of car collision by analyzing 800 videos of water escapes collected directly and training them in a man-made intelligence network.

So as to detect the purpose of collision, 'temporal information' to research the pattern of movement in consecutive frames and 'spatial information' to grasp the structure and shape of the item have to be analyzed concurrently, so the research team developed a 3D-CNN able to simultaneous evaluation. was used.

3D CNN (3D Convolutional Neural Networks) is a network based on CNN, a deep learning structure created by imitating the human optic nerve. The widely known 2D CNN deals with two-dimensional data resembling images, while the 3D CNN analyzes and learns from video by adding a time axis.

As a consequence of the character of a water escape accident through which the victim vehicle was identified, a preprocessing method was used to stop unnecessary background information from being input into the network by leaving a certain gap across the victim vehicle.

Vehicle crash images will be distinguished from movement patterns in non-collision situations because they show repetitive shaking movements during a collision.

The results of this research is that it is feasible to instantly check the movement of the item and the trail it took before and after a suspected water escape accident, which might significantly reduce work time in comparison with the present investigator in charge analyzing the video directly.

Subsequently, if this technology is applied to widely installed CCTVs, it might be used for crime prevention and evaluation, and is predicted to be highly effective in strengthening the protection of the community and stopping crime.

Professor Lee Yong-gu said, “This research consequence is most meaningful in that it significantly reduces the burden of analyzing vast CCTV images with advanced artificial intelligence technology,” adding, “We’ll improve social trust and safety by quickly identifying and processing accident situations through future commercialization.” “It is predicted that we’ll have the option to further increase it,” he said.

This research, led by Professor Yong-gu Lee and took part by Researcher In-woo Hwang, was conducted with support from the Ministry of Trade, Industry and Energy, the Ministry of Science and ICT, the Defense Acquisition Program Administration, and the Science and Public Safety Promotion Center. The research results were published online on February 19, 2024, within the renowned international academic journal 'JCDE (Journal of Computational Design and Engineering)'.

Reporter Oh Deok-hwan odh@aitimes.com