Computer Vision Development in Amsterdam

Logistics / Port Operations

A Rotterdam-based container terminal operator (processing 4.2 million TEU annually across 3 terminals, EUR 680M revenue, operating the most automated terminals in Europe) needed computer vision to automate container damage inspection, reduce disputes with shipping lines, and improve operational efficiency at terminal gates. (1) Gate throughput bottleneck: every container entering or leaving the terminal required damage inspection — a manual process where gate clerks visually inspected containers from 4 angles and recorded damage on paper forms. Gate inspection took an average of 90 seconds per container — creating congestion during peak periods when 200+ trucks queued at gates. Gate throughput limited terminal capacity. (2) Damage dispute cost: shipping lines and the terminal disputed container damage responsibility for 8.5% of movements — each party claiming the other was responsible for damage. Without objective, timestamped evidence, disputes were resolved through negotiation rather than fact — the terminal bore 60% of disputed damage costs. Annual damage dispute costs: EUR 12.8M. (3) Inspection consistency: 85 gate clerks across 3 terminals with varying levels of experience and attention. Damage detection rates varied from 72% to 95% between clerks. Night shift and rainy conditions reduced detection rates by an estimated 20%. Damage classification (minor/moderate/severe) was subjective — the same damage might be classified as moderate by one clerk and severe by another. (4) Crane damage: container damage occurring during crane operations (lift-on, lift-off, and stacking) was difficult to attribute. Without continuous visual monitoring, the terminal could not determine when during the handling cycle damage occurred — or whether the container was already damaged when it arrived. This ambiguity drove up insurance costs and complicated shipping line relationships. (5) Reefer inspection: refrigerated containers (reefers) required additional visual inspection — checking for damage to refrigeration units, power connections, and insulation integrity. Reefer damage could cause cargo loss worth hundreds of thousands of euros. Manual reefer inspection was particularly unreliable due to the complexity of the refrigeration equipment.

We developed and deployed a container inspection vision system over 12 weeks. (1) Gate inspection automation: high-resolution camera arrays installed at all terminal gates — 12 cameras per gate lane capturing all 6 container surfaces (top, bottom, 4 sides) plus doors, corner castings, and undercarriage. Capture: images captured automatically as trucks passed through the gate at driving speed (no stopping required for standard containers) — reducing gate processing to under 15 seconds per container. Damage detection: deep learning models trained on 2.8 million annotated container images detecting 28 damage categories per IICL (Institute of International Container Lessors) standards — dents, holes, cracks, corrosion, bent components, missing parts, and structural deformation. Severity classification: each detected damage item automatically classified by severity (cosmetic, operational, structural) with dimensional measurements derived from calibrated camera geometry. Container identification: OCR reading container number, owner code, and size/type code per ISO 6346 — linking damage records to specific containers automatically. (2) Timestamped evidence: every container movement through the terminal recorded with timestamped imagery — gate in, yard placement, crane operations, and gate out. Damage comparison: automated comparison of container condition between gate-in and gate-out — any new damage occurring during the terminal stay precisely identified and documented. Crane monitoring: cameras on STS (ship-to-shore) and RTG (rubber-tyred gantry) cranes capturing container condition during lifting operations — identifying the exact moment damage occurred during handling. Evidence package: automated generation of damage evidence packages including timestamped photos, damage classification, and dimensional measurements — providing irrefutable evidence for dispute resolution. (3) Reefer inspection: specialised camera and sensor system for refrigerated containers — visual inspection of refrigeration unit components (compressor, condenser, evaporator, controller), thermal imaging detecting insulation breaches and abnormal temperature patterns, and power connection integrity verification. Reefer health scoring: each reefer assessed with a health score based on visual and thermal analysis — containers with low scores flagged for inspection by reefer technicians before loading. (4) Analytics platform: real-time dashboard showing damage rates by shipping line, vessel, origin port, and container age — identifying patterns (specific shipping lines with higher damage rates, specific vessels causing damage during loading, containers from specific origins arriving damaged). Predictive maintenance: analysis of terminal equipment (crane spreaders, twist locks, and truck chassis) identifying equipment contributing to container damage — enabling proactive maintenance before damage incidents. Cost analytics: automated calculation of damage liability — attributing costs to the responsible party (terminal, shipping line, trucker) based on timestamped evidence. (5) Integration: TOS integration (Terminal Operating System — damage records automatically linked to container movements, vessel visits, and billing records), EDI messaging (automated damage notifications to shipping lines in COPARN/BAPLIE format), and customs integration (damage documentation provided to customs authorities when required for inspection or duty assessment).

Gate throughput: inspection time from 90 seconds to under 15 seconds per container (driving through without stopping). Gate capacity increased by 35% — eliminating peak period queuing. Truck turnaround time reduced by 18 minutes on average — improving driver satisfaction and terminal service reputation. Damage detection: detection rate from variable (72-95% depending on clerk) to 96.5% consistent across all conditions — day, night, rain, and fog. 28 damage categories classified per IICL standards with 94% classification accuracy. Zero detection variation between shifts or weather conditions. Dispute resolution: damage disputes reduced from 8.5% to 1.2% of movements — timestamped evidence providing objective attribution. Disputed movement cost: from EUR 12.8M to EUR 1.8M annually (EUR 11M savings). Dispute resolution time: from 45 days (negotiation-based) to 5 days (evidence-based). Crane damage attribution: specific equipment items (2 crane spreaders and 4 truck chassis) identified as causing disproportionate damage — maintenance and replacement reduced crane-related damage by 62%. Reefer: reefer inspection accuracy improved — 3 reefer failures prevented in the first year (estimated cargo save: EUR 850K). Reefer health scoring enabled proactive maintenance, reducing reefer breakdown rate at the terminal by 45%. Staffing: gate clerks from 85 to 25 (retained for exception handling, manual verification, and system monitoring). Annual value: EUR 11M (dispute reduction — damage dispute costs from EUR 12.8M to EUR 1.8M), EUR 8.5M (gate efficiency — increased throughput enabling additional container handling without terminal expansion), EUR 4.2M (staffing — 60 gate clerk positions redeployed through attrition and retraining), EUR 2.8M (equipment optimisation — reduced damage incidents through predictive maintenance and equipment identification), EUR 1.2M (reefer protection — prevented cargo losses and reduced reefer breakdown) = EUR 27.7M. Development cost: EUR 380,000. Payback: under 6 days.