Sumitomo and NEC AI near-miss analysis: safety and data lessons for site engineers
Reviewed by Tom Sullivan

First reported on The Construction Index
30 Second Briefing
Sumitomo Heavy Industries and NEC are jointly developing an AI and computer vision system that uses camera feeds from hydraulic excavators and SHI’s SHICuTe ICT/IoT platform data to automatically detect “risk scenes” and generate structured near-miss reports. NEC’s 2023 video recognition and generative AI technology, previously used for road traffic accident analysis, will fuse time- and location-stamped video with machine operating logs as multimodal data to characterise hazardous and prohibited behaviours. Following a successful proof of concept in September 2025, full development starts April 2026, with global deployment targeted for broader construction-site safety management.
Technical Brief
- Extraction AI model is trained on SHICuTe platform datasets from real hydraulic excavator operations.
- Cameras are mounted directly on hydraulic excavators, providing machine-eye video of worker–plant interactions.
- NEC’s multimodal engine stores risk scenes with explicit temporal and spatial metadata for replay and auditing.
- Hazardous and prohibited behaviours are defined using past accidents, equipment failures and “operations requiring particular attention”.
- Company-specific rulesets are incorporated, enabling alignment with internal safety protocols and method statements.
- September 2025 technical proof-of-concept confirmed automatic extraction of near-miss cases from excavator-mounted video alone.
- Joint development from April 2026 targets expansion of detectable near-miss types and richer narrative report content.
- Longer term scope includes unsafe site conditions not obvious to workers, extending beyond direct worker–machine conflicts.
Our Take
Safety-tagged software items in our database increasingly sit alongside project-delivery stories, so embedding AI near-miss analysis into SHI Group’s project workstreams could influence how UK clients structure contractual safety KPIs and reporting obligations from the mid-2020s onward.
Prepared by collating external sources, AI-assisted tools, and Geomechanics.io’s proprietary mining database, then reviewed for technical accuracy & edited by our geotechnical team.
Related Articles
Related Industries & Products
Construction
Quality control software for construction companies with material testing, batch tracking, and compliance management.
Mining
Geotechnical software solutions for mining operations including CMRR analysis, hydrogeological testing, and data management.
QCDB-io
Comprehensive quality control database for manufacturing, tunnelling, and civil construction with UCS testing, PSD analysis, and grout mix design management.

