NEC has newly developed AI that uses a vision-language model to score site conditions from live camera footage and announced technology capable of detecting disaster precursors, such as river overflow, flooding and heavy snowfall, at an early stage. Practical implementation is targeted during FY2027.
NEC announced that it had newly developed AI that uses a vision-language model (VLM) to score site conditions from live camera footage and detect at an early stage precursors that could lead to disasters. The technology uses existing cameras owned by local governments and infrastructure operators, helping to speed up on-site checks for disaster prevention and response decisions.
To overcome the difficulty of detecting subtle changes, identified as a limitation of conventional VLMs, the technology analyzes internally calculated confidence levels and uses them to score site conditions numerically. Prompts are generated automatically to set analysis conditions suited to each site, reducing the burden of deployment and operation.
The technology can capture changes in site conditions over time as continuous numerical values, enabling detailed assessment of factors such as flood depth and snow accumulation. It also automatically generates prompts suited to the subject being monitored and the composition of the footage, automatically determines the areas on which the VLM should focus, and stabilizes scoring. NEC aims to improve the efficiency of monitoring operations conducted by local governments and infrastructure operators and enhance regional safety.
The value offered by the technology includes detecting disaster precursors at an early stage and speeding up on-site checks and response decisions. It is expected to help prevent disasters before they occur and limit their expansion. NEC says that automatically generating prompts suited to each site will reduce the trial-and-error burden on technical specialists.