An independent magazine about AI at work in construction. Every story starts on a real job site, with the people who used the technology and the results they will answer for.
Kajima's Camera Gantry Cuts Osaka Rebar Cage Witness Checks to 100 Minutes | ConstructionMagazine.ai
Kajima's Camera Gantry Cuts Osaka Rebar Cage Witness Checks to 100 Minutes
Kajima and Mitsubishi Electric Engineering put two AI stereo cameras on a gantry for an Osaka rail tunnel wall. Kajima's unaudited figures (11 Sep 2026): owner witness time per cage fell from about 300 to about 100 minutes.
Kajima says a camera gantry that rolls along a reinforcement cage cut the time its client's inspector spent witnessing each cage from about 300 minutes to about 100 on an Osaka rail tunnel job. Nikkei xTECH carried the story on 1 October 2026, three weeks after Kajima announced the system on 11 September with its development partner, Mitsubishi Electric Engineering of Minato, Tokyo.
This account reflects the sources as of 12 October 2026. The project, the client, the joint venture and the cage dimensions come from Kajima's Japanese press release. The time savings are Kajima's numbers, unaudited, and the trade outlets that covered the story restate them from the release rather than from their own measurement. No one is quoted by name anywhere in the coverage.
The first job is a diaphragm wall on Osaka's Naniwasuji Line
The system went first onto the main civil works for the Naniwasuji Line's Nankai Shin-Namba branch tunnel and the Sennichimae-dori shield tunnel in Osaka, according to the release. Kansai Rapid Railway Co. is the client. A joint venture of Kajima, Konoike Construction and Tobishima Corp. holds the contract, which runs from April 2025 to March 2029.
The work includes a reinforced concrete diaphragm wall, the structure US engineers often call a slurry wall, used to hold back soil and cut off groundwater around a deep excavation. Kajima's release lists 24 wall elements built from 30 reinforcement cages. Each cage is assembled in five lots, runs 54.2 m in total length and is up to 2.7 m wide.
The release gives bar sizes from D16 to D51, the Japanese designations for deformed bars of roughly 16 mm to 51 mm nominal diameter. Main bars are spaced 150 to 300 mm apart and distribution bars at 150 or 200 mm.
Why an owner's inspector watches the rebar before the pour
On Japanese public and infrastructure work, the owner's representative witnesses the rebar check before concrete is placed. The contractor measures the bars against the drawings while the client's inspector watches and accepts the result. IBTimes JP describes the step as the way contractors and owners confirm that reinforcement placement meets the design before the pour. Kajima's technology page says the check takes place under owner witness at every concrete cycle.
For a US or UK reader, the nearest equivalent is the pre-pour rebar inspection by the engineer's or owner's representative. Kajima's technology page describes the traditional routine: markers on the bars to tell sizes apart, scale staffs laid against the steel to show spacing, and a handwritten blackboard photographed with the result.
11 September 2026. Primary source: system design (two stereo cameras, one tablet, gantry unit), joint development with Mitsubishi Electric Engineering, Naniwasuji Line project details (client, JV, term, 30 cages, five lots, 54.2 m, 2.7 m, D16 to D51), lot-by-lot inspection, the 300 to 100 minute and about 70% figures (company figures, unaudited), reduced staff, no walking on cages, the 2023 MLIT draft guideline.
12 September 2026. English rendering of the release, including project and JV names; supports the description of witness inspection as how contractors and owners confirm placement meets design before the pour. Restates Kajima's figures.
25 September 2026. Japanese trade coverage restating the release, including Kajima's use of AI rebar inspection on bridge superstructures and the 300 to 100 minute figures.
In 2023 Japan's Ministry of Land, Infrastructure, Transport and Tourism (MLIT) issued a draft guideline for measuring as-built rebar with digital data. Kajima's technology page says the guideline allows a digital camera of specified performance to check bar diameter, count and spacing. Kajima's release says it has since used AI rebar inspection on many concrete structures, including bridge superstructures it built.
Deformed bars of the kind checked for count, spacing and diameter. Stock photo from Unsplash, not the Osaka site.
Two stereo cameras ride a gantry along the cage
The new setup builds on Kajima's earlier portable AI rebar inspection system, which pairs one stereo camera with a tablet and is carried by an inspector. The mobile version mounts two stereo cameras on a gantry-type frame that travels along the cage. A single tablet controls both cameras and merges what they capture, so the whole length of a long cage can be measured as the frame moves, per the release.
Kajima's release explains why the handheld tool struggled on diaphragm-wall cages. Inspectors had to follow the assembly as the bars went in, moving and photographing as they went, which the release says took an enormous amount of time.
Wide, multi-tier cages also left blind spots for a single stereo camera. Nikkei xTECH's lead paragraph attributes the blind spots to the cages' stiffening frames and says using two sets of stereo cameras removes them.
Site layout added a further burden. Where there is no room to assemble cages beside the wall, they are prefabricated elsewhere and split for transport, which multiplies inspection points and travel for both the owner's and the contractor's staff, the release says. It does not say whether the Naniwasuji cages were assembled on site or away from it.
Nikkei xTECH reports that the system checks bar count, spacing and diameter from the captured images. The release includes a picture of the inspection data sheet the system produces but does not describe what fields it records.
How a diaphragm-wall cage was inspected lot by lot, as Kajima's 11 Sep 2026 release describes it
1Rebar crew assembles the lower bars of one lot (each cage has five lots)
→
2Gantry travels along the lot while two stereo cameras shoot and measure continuously
→
3One tablet merges both cameras' data and checks bar count, spacing and diameter
→
4Owner's inspector witnesses the check on the assembled section
→
5Crew assembles the upper bars and the gantry repeats the pass
→
6Inspection moves to the next lot until all five are done
→
7System outputs an inspection data sheet (sign-off method not described in the release)
Kajima inspected each cage lot by lot as the crew built it
On the Osaka wall, Kajima photographed and measured each of the five lots of a cage separately. The inspection followed the rebar crew, lower bars first and upper bars once they were in place, so that checking happened on whatever had just been assembled, according to the release.
Kajima says this staged routine prevented schedule delays caused by waiting for inspection. The release gives no figure for time saved across the program, only the per-cage witness time.
The time savings are Kajima's figures
Owner witness inspection time per rebar cage, approximate minutes (source: Kajima press release, 11 Sep 2026; company figures, unaudited)
Kajima's release says owner witness inspection time per cage fell from about 300 minutes with conventional scale-based methods to about 100 minutes with the mobile system. It describes that as a cut of about 70%. These are the company's numbers, unaudited. IBTimes JP, Digicon and BUILT repeat them from the release, and none reports an independent measurement.
On the two rounded figures, going from 300 minutes to 100 is a reduction of two-thirds, about 67%. Kajima's figure of about 70% may rest on more precise underlying times, but the release does not publish them. It also does not say whether 100 minutes is an average across the 30 cages, a typical cage or a single trial.
Kajima also says the job needed fewer inspection staff. The release does not say how many people the check took before or after, or whether the reduction fell on the client's side, the joint venture's side, or both.
The release adds a safety claim: inspectors no longer have to walk on top of the cages. Digicon put that point in its headline. By Kajima's account, the measuring now happens from the gantry rather than from the steel.
The accuracy numbers belong to the handheld system
Kajima's technology page reports a blind test led by MLIT's National Institute for Land and Infrastructure Management, in which the system scored 100% on bar spacing and 94% on bar diameter. Those results are for the earlier portable system. Kajima has published no accuracy figures for the mobile gantry version.
The same caution applies to adoption. The portable system is registered in MLIT's NETIS new technology database as KT-230164, and the release says more than 300 companies have adopted it. That count is for the portable tool. The mobile system has one reported deployment, on this Osaka job.
The technology page says the portable system meets MLIT inspection criteria when shooting from 1 to 2 m at an angle of 30 degrees. It also notes that telling apart deformed bars one size different has been hard for similar technologies. Neither the release nor the trade coverage says whether the gantry keeps its cameras inside that window along a cage up to 2.7 m wide.
What the sources leave out
The release does not say how many inspectors the crew dropped to, or from what number. It gives no cost for the gantry, the cameras or the software, and it does not say whether the joint venture bought or leased the equipment.
It does not report accuracy for the two-camera setup. It also does not say how many of the 30 cages had been inspected with the system by the time of the announcement.
The largest gap is the sign-off. The release does not explain how Kansai Rapid Railway's inspector accepts the digital record: whether the tablet's data sheet becomes the inspection record, whether bars are still spot-checked with a tape, or whether the inspector watches the gantry pass live.
Nikkei xTECH's full article sits behind a paywall, and only its lead paragraph was read for this story, so it may carry details not reflected here. Mitsubishi Electric Engineering's role is described only in Kajima's materials.
An X post by @SiteSpecs on 3 October cited the same 300 and 100 minute figures. In the magazine's paraphrase, the post argued that the inspector's signature stays but the time around it shrinks. That is commentary resting on Kajima's numbers, not separate reporting.
What to ask before putting a camera gantry on your cage line
Start with the owner. Ask the client's inspector, in writing, whether they will accept a camera-measured record in place of a witnessed tape check, which spot checks they still want, and who keeps the data. On public work in Japan the 2023 MLIT draft guideline gives that conversation a reference point. Elsewhere you may need to write the method into the inspection and test plan before the first cage.
Ask the vendor for accuracy on your configuration. A blind-test score for a handheld camera at 1 to 2 m does not tell you how two cameras on a moving frame perform on a wide, multi-tier cage with D51 bars. Ask for results on bars one size apart, since Kajima's own technology page flags that as the hard case.
Ask for the baseline behind any time claim. Kajima's 300 minutes was its conventional method on this wall. Time your own witness inspection per cage before a trial, so the before-and-after comparison is yours rather than the supplier's.
Ask where the cages will be built. The gantry needs a clear run along the full cage, 54.2 m on the Osaka job. If your cages are prefabricated off site and split for transport, find out whether the gantry travels with them or the inspection moves to the yard.
Ask who still has to climb onto the steel. If the answer is nobody, revise the fall-protection plan and the inspection plan together, because the safety gain Kajima reports depends on the measurement staying on the ground.
· Nikkei xTECH (日経クロステック)
1 October 2026. Newest dated coverage; paywalled, lead paragraph only read. Supports the gantry with two sets of stereo cameras removing blind spots from stiffening frames, and AI checking bar count, spacing and diameter from images.
Undated; accessed 5 October 2026. Vendor page for the earlier portable system only: owner-witnessed checks at each concrete cycle, markers and scale staffs in the traditional method, the MLIT guideline's allowance for digital cameras, the NILIM-led blind test (100% spacing, 94% diameter), the 1 to 2 m and 30 degree shooting conditions, and the difficulty of one-size-apart bar diameters. None of these figures apply to the mobile gantry system.
3 October 2026; captured in the 5 October 2026 X Latest ingest. Secondary commentary only; cites Kajima's 300 to 100 minute figures and argues the inspector's signature remains while the time around it shrinks. Used as paraphrase, not as a source of fact.