The system will draw on SHI鈥檚 expertise in construction machinery and data analysis, combined with NEC鈥檚 video recognition and generative AI capabilities. Work on building the system is due to start next month, April 2026.
The aim is to spot near-misses when they happen, and generate useful reports to enhance site safety. It will make use of information drawn from camera footage, alongside data from SHI鈥檚 hydraulic excavators.
The new system will make use of an extraction AI model, trained on real world hydraulic excavator data accumulated on SHICuTe, SHI Group's ICT/IoT common platform. This will spot potentially hazardous scenes, dubbed 鈥榬isk scenes鈥�, using site-specific and time-dependent video footage and work logs.
These risk scenes, together with operational data from the hydraulic excavators, will then be analyzed using NEC's proprietary technology that combines video recognition with generative AI, and stored as multimodal data incorporating temporal and spatial information. NEC first in 2023, using it to generate detailed reports from video recordings of road traffic accidents.
Based on this data, along with SHI's expertise in construction-site machinery operations and human workflows, the system will cross reference hazardous and prohibited behaviour data. These will be defined by accidents, construction equipment failures, and operations requiring particular attention. They will be combined with company specific data. Based on these matching results, the system will automatically identify the risk scenes that should be reported and automatically generate high quality near-miss reports that provide a concise summary of the circumstances surrounding each incident.
The two companies started work on the project in September 2025, conducting a technical proof of concept. that automatically extracts near-miss聽incidents and generates reports from video footage captured by cameras聽mounted on hydraulic excavators. The results confirmed that, based on the聽risk scenes extracted from the footage, the system was able to report near-miss cases, including potential accident scenarios and their associated聽circumstances.
Building on these outcomes, the joint development starting聽in April 2026 will aim to further expand the types of near-miss incidents聽that can be identified, and to enhance the report generation capabilities. The goal will be to support customer safety management needs, and overall site safety. Longer term, the system will expand beyond interaction between workers and machinery that may lead to occupational accidents, to include unsafe conditions that may not be readily recognised by workers. The goal is to offer the system to customers around the world.