Being explored. Feasibility is not yet proven.
Visible damage
Possible cracks, dents, corrosion and leaks, where image quality and access allow.
SIWOM Safety
Being explored. Feasibility is not yet proven.Hazard flags, reviewed by people.
SIWOM Safety is research into camera-assisted inspection and hazard flagging for ports, yards and industrial sites. AI flags possible hazards; a qualified person reviews every flag and decides.
Safety-first philosophy
AI assists. People decide.
Industrial safety depends on trained people, good procedures and proper inspections. SIWOM Safety explores whether AI can help those people notice potential hazards sooner and record what they find more consistently.
It is never a replacement for them. AI flags potential hazards for qualified human review. It does not certify a site as safe, and SIWOM does not control cranes, vehicles or dangerous machinery.
Workflow
Observe, detect, review, respond, improve.
Step 1: Observe
Cameras and suitable sensors capture the scene.Step 2: Detect
Software flags a potential hazard, with a confidence band.Step 3: Review
A qualified person checks the flag and decides.Step 4: Respond
People act: inspect, pause work, escalate or dismiss.Step 5: Improve
Outcomes feed back to reduce false alarms over time.
Inspection demo
You are the reviewer.
Pick a scenario, select a flagged detection and decide: confirm, mark as false alarm or escalate. Each decision is logged.
Illustrative demo data : not connected to live operations
Loading interactive demo…
CAM-0714:12:08 IST · still · demo
Real reference photograph. Review areas and confidence labels are scripted demo examples, not AI findings.
Detections in this frame · illustrative
HighPending review
Visible condition · hardware review
Latch condition review
A manually selected area around the container latch. A reviewer would inspect its condition; the image alone does not establish a defect.
- Location
- Yard B, Bay 4 — demo
- Camera
- CAM-07
- Captured
- 18 Oct 2026, 14:12:08 IST
- Confidence
- Moderate (illustrative)
Recommended next step
Ask a qualified inspector to assess the latch and record the outcome before deciding on further action.
A qualified reviewer decides. Choose one:
4 detections pending review · demo
| Time | Detection | Action | By |
|---|---|---|---|
| 14:05:47 IST | DET-2037 · Person near reversing truck | Escalated | Sent to shift supervisor (demo reviewer) |
| 13:58:12 IST | DET-2033 · Possible spill | Dismissed | Reflection on wet ground (demo reviewer) |
About these detections
Real photographs with manually authored sample review areas; no AI has analysed these images and no actual violation is asserted. Real-world performance depends on camera quality, sensor coverage, lighting, weather and validation. AI flags potential hazards for qualified human review; it does not certify a site as safe or assess hidden structural, electrical or material defects.
Computer vision examples
Being explored. Feasibility is not yet proven.What we are exploring.
- Being explored. Feasibility is not yet proven.
PPE compliance
Possible missing helmets or hi-vis vests in areas where they are required.
- Being explored. Feasibility is not yet proven.
Restricted-zone intrusion
People or vehicles inside marked zones such as exclusion areas.
- Being explored. Feasibility is not yet proven.
Lifting and traffic conflicts
People near active lifting zones, and vehicle–pedestrian conflict points.
- Being explored. Feasibility is not yet proven.
Load monitoring
Load conditions, only with suitable sensors. Cameras alone cannot weigh a load.
Feasibility first
Each of these is a research question, not a capability. Whether it works on a given site depends on camera quality, sensor coverage, lighting, weather and validation on that site's data.
PPE and restricted zones
Being explored. Feasibility is not yet proven.Flags for the right person, quickly.
A possible missing helmet or a person near an active lifting zone matters only if the right person hears about it in time. The concept is simple: a flag with the frame, location, camera and time goes to a reviewer, who decides what happens next.
- Zones are drawn and owned by the site's safety team
- Confidence is shown as a band (low, moderate, high), never a precise score
- A possible PPE gap is a prompt to check, not a disciplinary record
Incident management
Being explored. Feasibility is not yet proven.From observation to corrective action.
- Being explored. Feasibility is not yet proven.
Inspection records
Consistent records with photos, location and the inspector's findings.
- Being explored. Feasibility is not yet proven.
Incident reporting
Structured reports that capture what happened, where and who responded.
- Being explored. Feasibility is not yet proven.
Corrective actions
Actions with owners and due dates, tracked to closure.
Human review and escalation
Every flag reaches a person.
Reviewers can confirm a hazard, mark a flag as a false alarm or escalate it to a supervisor. Each choice is time-stamped in an audit trail, so the record shows who decided what, and when.
- Confirm hazardRecorded as a confirmed observation; follow-up actions start.
- Mark as false alarmRecorded with the reason, and used to reduce future false alarms.
- Escalate to supervisorSent to the person with authority to pause work or act.
Limitations and responsible use
What this cannot do.
These limits are part of the product, not fine print.
Cameras see surfaces only
Ordinary cameras cannot assess internal structural integrity, electrical safety or hidden material defects. Those need engineering inspections, specialised sensors and non-destructive testing.
No guarantees
AI flags potential hazards for qualified human review. It never guarantees a site is safe, and detection performance depends on cameras, coverage, environment and validation.
No control of machinery
SIWOM does not control cranes, vehicles or dangerous machinery. Stopping work is a human decision made through the site's procedures.
Validated before relied on
Any deployment would be validated on the site's own data, with false-alarm and miss rates shared openly with the safety team.
Visible surface conditions, where image quality allows: possible dents, corrosion, leaks, missing PPE or people inside marked zones. Ordinary cameras cannot assess internal structural integrity, electrical safety or hidden material defects. Those need engineering inspections, specialised sensors and non-destructive testing.
Performance depends on camera quality, sensor coverage, lighting, weather, viewing angle and validation on each site's own data. Expect false alarms and missed detections. Any deployment would be validated on site before people rely on it.
No. It flags potential hazards for qualified human review. People decide what happens next and remain accountable. It never certifies a site, an asset or a lift as safe.
No. SIWOM does not control cranes, vehicles or dangerous machinery. Any stop or pause is a human decision made through the site's existing procedures.
Not yet. SIWOM Safety is in research. We are looking for industrial partners to explore it with on real sites, under their safety processes.
Our approach is to flag conditions, not identify people: no face recognition, footage access limited by role, and retention periods agreed with each site. These are design intentions, not certifications.
FAQ
Questions we hear.
No. It is in research. We are looking for partners who want to explore it on real sites, under their existing safety processes.
Possibly not. Existing CCTV can sometimes be used, but coverage, resolution, angle and lighting decide what is feasible. That is assessed per site before anything is promised.
No. It can make inspection records more consistent and easier to audit, but it does not certify a site, asset or process as safe or compliant. Qualified people and the applicable inspections still do that.
Reviewers mark them as false alarms, and that outcome is logged. Over time, those outcomes help tune the system for the site. False alarms will never reach zero.