Computer Vision Development in Miami, FL

Port Logistics / Terminal Operations

A terminal operator at PortMiami (handling 420,000 TEUs annually, 180 employees, serving 8 ocean carrier customers) was operating manual container inspection at terminal gates. Gate inspection process: every container entering or exiting the terminal required visual inspection — gate checkers examining each container for damage (dents, holes, rust, structural deformation), verifying seal integrity (intact seal with correct seal number), and recording container number and ISO size/type code. Inspection staff: 12 gate checkers working across 4 gates in 3 shifts (24/7 operation). Average inspection time: 2.8 minutes per container — the time creating gate queues during peak periods and limiting terminal throughput. The bottleneck: during peak hours, the gate queue could extend to 45+ minutes — trucks idling, drivers frustrated, and terminal capacity constrained by gate throughput rather than yard or vessel capacity. Damage dispute resolution: 14% of damage claims disputed between the terminal and ocean carriers or trucking companies — disputes arising because the manual inspection process was subjective (different gate checkers having different damage thresholds), incomplete (not every angle of every container examined), and poorly documented (damage noted in text descriptions rather than photographic evidence). Annual disputed damage claims: $1.8M — with the terminal absorbing approximately $680K in claims that should have been attributed to arriving carriers. Data entry errors: container number recording accuracy was 96.8% — meaning 3.2% of containers entering the terminal had incorrect numbers in the system. The errors: causing container location confusion, delayed pick-ups, and customer service issues. Each mis-recorded container: consuming an average of 35 minutes of staff time to locate and correct. Seal verification: manual seal checking was unreliable — gate checkers visually confirming seal presence but not consistently verifying seal numbers against shipping documentation. Seal discrepancies: discovered during later audits rather than at the gate, creating security and compliance issues.

We developed and deployed a container gate computer vision system over 12 weeks. Camera infrastructure (weeks 1-4). Gate camera arrays: each of the 4 terminal gates equipped with 8 high-resolution cameras — capturing every container from all angles (front, rear, left side, right side, top, and undercarriage) as it passes through the gate at normal speed. Camera specifications: 12MP industrial cameras with global shutter (eliminating motion blur at gate transit speed), IR illumination (enabling night operation), and IP67-rated enclosures with active cooling (handling Miami's heat and humidity). Trigger system: inductive loop sensors in the gate road surface triggering the camera array as each container chassis enters the capture zone — ensuring consistent positioning and timing for every capture. Image quality: calibration system ensuring consistent image quality across all conditions — daytime sun, nighttime, rain, fog, and the dramatic lighting changes during Miami's frequent tropical thunderstorms. Container identification (weeks 3-7). OCR model: AI reading container numbers and ISO size/type codes from captured images — trained on 150,000+ container images captured during a 6-week pre-deployment data collection phase. Accuracy: 99.4% container number reading accuracy on validation data — 2.6 percentage points better than manual recording. Multi-angle reading: the system reading container numbers from multiple camera angles — if the primary side image is obscured (by mud, damage, or poor paint), the system automatically using the rear or other visible markings. Real-time verification: container numbers verified against the terminal's gate appointment system — confirming that the container arriving at the gate is the one expected. Mismatches: flagged immediately for gate staff intervention. Damage detection (weeks 5-10). Damage classification model: AI identifying and classifying container damage — trained on 85,000 annotated container images with damage categories including dents (minor, moderate, severe), corrosion (surface, structural), holes (patching needed, condemned), structural deformation (rail damage, corner casting damage, door alignment), and floor damage (visible through undercarriage camera). Severity scoring: each detected damage item scored on a 1-5 severity scale — level 1-2 (cosmetic, no operational impact) processed automatically, level 3-4 (significant, requiring monitoring or repair) flagged for operator review, level 5 (potentially condemned, safety risk) triggering automatic gate hold. Photographic evidence: every damage detection linked to annotated images — showing the exact location, type, and severity of damage with measurement overlays. The evidence: creating an objective, timestamped record for damage dispute resolution. Before/after comparison: containers exiting the terminal compared against their arrival images — any new damage identified and attributed to terminal operations, creating accountability for terminal damage. Seal verification (weeks 7-12). Seal detection: AI identifying container seals in gate images — detecting seal presence, reading seal numbers through OCR, and assessing seal condition (intact, broken, cut, or missing). Seal number verification: captured seal numbers compared against electronic advance manifest data — verifying that the seal on the container matches the seal number declared by the shipping line. Discrepancies: flagged immediately for gate staff intervention before the container enters the terminal. Tamper detection: AI identifying visual indicators of seal tampering — replacement seals with different colour or type than declared, seal housing damage suggesting removal and replacement, and inconsistencies between seal type and container age/condition.

The gate CV system deployed across all 4 gates within 14 weeks. Gate throughput: container processing time: from 2.8 minutes (manual inspection) to 18 seconds (drive-through CV capture with automated processing). Gate throughput: from 21 containers per gate per hour to 120+ containers per gate per hour — the gate bottleneck eliminated entirely. Peak queue time: from 45+ minutes to under 5 minutes — driver satisfaction improved significantly. The throughput improvement: enabling the terminal to handle 15% more volume without adding gate infrastructure. Container identification: number reading accuracy: from 96.8% (manual) to 99.4% (CV) — a 81% reduction in errors. Misidentified containers per month: from 1,120 to 210 — and the CV system catching its own errors through multi-angle verification. Staff time savings from reduced mis-identification: 28 hours/week eliminated. Damage management: damage detection rate: 94.2% of significant damage (level 3+) identified automatically — compared to estimated 72% identified by manual gate checkers (the remaining 28% either missed or subjectively categorised below reporting threshold). Damage dispute resolution: from 14% of claims disputed to 3.8% — photographic evidence with timestamps and severity scores providing objective documentation that both parties accepted. Disputed claim value: from $1.8M annually to $490K — and terminal recovery of incorrectly absorbed claims improving by $420K/year. Before/after comparison: terminal now identifying and documenting damage occurring during terminal operations — reducing terminal-caused damage by 28% through accountability (operations staff aware that handling damage is now tracked and attributed). Seal verification: seal discrepancies caught at gate: 34 in the first 6 months (vs. 0 caught at gate previously — all discovered during later audits). The 34 discrepancies: including 3 genuine security concerns escalated to CBP, 18 documentation errors corrected before entry, and 13 seal replacements properly documented. Security compliance: the terminal's CBP compliance rating improving based on enhanced seal verification capability. Financial results: throughput improvement value: $2.4M/year (15% volume capacity increase at existing infrastructure). Damage dispute savings: $1.31M/year (reduced disputes + recovered incorrectly absorbed claims). Gate staff optimisation: $680K/year (12 gate checkers reduced to 4 — 8 redeployed to yard operations). Data accuracy improvement: $180K/year (reduced time correcting container identification errors). Total annual impact: $4.57M/year. Computer vision system investment: $480,000. Annual operations (including hardware maintenance): $96,000/year. First-year ROI: 852%.