{"resourceId":"tmb-ai-bus-passenger-counting-cameras-2026-11-04","versions":[{"version":"external-4c8f520d2e78fbaed2a3aebb2381c7d40a474312fb79a67f9e91626fa10ed2b5","resource":{"id":"tmb-ai-bus-passenger-counting-cameras-2026-11-04","title":"Barcelona TMB uses AI on 4,500+ bus cameras for real-time passenger counts and camera health","organization":"Transports Metropolitans de Barcelona (TMB)","sector":"Transit","geography":"Barcelona, Spain","publishedAt":"2026-11-04","publicationDate":"2026-11-04","eventDate":null,"sourceName":"Barcelona City Council Press Service","sourceLabel":"Municipal press release / agency announcement","sourceUrl":"https://ajuntament.barcelona.cat/premsa/2026/11/04/els-busos-de-tmb-compten-el-passatge-de-cada-vehicle-en-temps-real-per-millorar-el-servei/","evidenceClass":"independent-reporting","outcomeClass":"emerging","topics":["infrastructure","operating-model","data-security"],"finding":"At Tomorrow Mobility World Congress 2026, TMB said it now has real-time passenger counts per bus by applying AI to images from more than 4,500 onboard CCTV cameras covering the full bus fleet, fused with validations, surveys, occupancy alarms and video counting. The in-house bus technology team built the system with AI vendors; models are audited and continuously retrained. Outputs are intended for service regulation and passenger information, and the same stack flags failed or misaimed cameras to speed maintenance. A related bus-lane intrusion detection pilot also uses AI on cameras.","sledRelevance":"A fleet-wide reuse of existing transit CCTV for occupancy AI, with explicit audit/retrain loops and a dual use for camera maintenance—directly comparable to U.S. transit APC modernization.","evidence":"Official city press note; no published accuracy rates, latency figures or before/after reliability metrics.","architectureImplications":"Computer vision on fleet CCTV fused with fare validations and other onboard occupancy signals; camera health monitoring is a second output of the same stack.","governanceImplications":"Continuous result audits and retraining are cited as quality controls; passenger-image retention and purpose limits are not described.","securityPrivacyImplications":"Video of riders is processed for counts; privacy, retention and access controls are not specified in the release.","caveats":"European operator; Catalan-language primary source; measured accuracy not disclosed; passenger-info use is planned rather than proven in the note.","streamIds":["transportation"],"roles":{"sales":"Sales framing: TMB turned an existing 4,500-camera bus CCTV plant into real-time occupancy AI plus camera-health detection, with ongoing audit and retrain loops. For U.S. transit agencies, the opening is APC modernization on cameras already installed—not a greenfield sensor buy. Ask how today APC accuracy and camera uptime are measured. Interpretation; no accuracy numbers are published.","engineering":"Engineering focus is multi-signal fusion: vision counts plus validations, surveys, alarms and video counting, with continuous model audit/retrain and camera-fault detection. Pre-sales engineers should map CCTV coverage, edge vs. central inference, and PII handling for rider video. The release names no model or vendor stack detail. Interpretation.","delivery":"Delivery work is standing up occupancy quality monitoring, service-regulation use cases and camera-maintenance tickets from the same AI. Bus technology owns the stack; operations consumes counts. Acceptance criteria should be proposed as count accuracy vs. manual audits and mean time to fix bad cameras. Interpretation; outcomes are not yet published."},"roleDetails":{"sales":{"customerProblem":"Transit operators lack reliable real-time per-vehicle occupancy and struggle to keep large CCTV fleets healthy.","stakeholders":["Transit COO / service planning","Bus technology / ITS lead","Customer information lead","Maintenance"],"discoveryQuestions":["What APC accuracy and latency do you have today?","How many onboard cameras and what is their uptime?","Who may use rider video for analytics?"],"valueHypothesis":"Vision AI on existing CCTV fused with fare data may improve occupancy estimates and camera maintenance; unproven quantitatively here.","potentialEngagement":"Bounded corridor pilot comparing AI occupancy to manual counts and APC.","unsupportedClaims":["Published accuracy rates","Service reliability gains","Passenger information quality improvements"]},"engineering":{"fit":"Fits agencies with dense onboard CCTV and digitized fare validations.","architectureIntegration":"Vision models on fleet video fused with validation and alarm feeds; camera health as secondary signal; details not disclosed.","prerequisites":["Fleet CCTV access","Validation data feed","Model audit process"],"constraints":["Privacy rules for rider video","European regulatory context","Vendor collaboration not specified"],"security":"Segment video access, minimize retention of identifiable frames, log model versions and audit results.","proofOfValue":["Manual occupancy audit sample","Camera-fault detection precision/recall","Service-regulation decision latency"]},"delivery":{"work":"Deploy counting pipeline, audit cadence, ops dashboards and maintenance tickets.","dependencies":["Bus tech team","AI vendors","Operations and maintenance"],"ownership":"TMB bus technology team with operations consumers.","skillsAdoption":"Ops training on using real-time occupancy; maintainers on camera-fault tickets.","governanceCheckpoints":["Model audit before production use","Privacy review of video analytics","Retrain trigger policy"],"acceptanceCriteria":["Proposed: occupancy MAPE vs. manual audits","Proposed: camera-fault MTTR reduction","Proposed: documented retrain log"],"risks":["Privacy backlash","Bias in crowded/lighting conditions","Over-trust without published accuracy"]}},"retrievedAt":"2026-11-05T22:09:09-06:00","enrichedAt":"2026-11-05T22:12:57-06:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Real-time occupancy can improve crowding information for riders; camera-maintenance workload shifts toward AI-flagged faults.","procurementImplications":"In-house build with expert AI vendor collaboration; contract terms not disclosed.","operatingModelImplications":"Occupancy AI feeds service regulation and passenger info; same stack supports camera maintenance.","sourceVerification":{"openedUrl":"https://ajuntament.barcelona.cat/premsa/2026/11/04/els-busos-de-tmb-compten-el-passatge-de-cada-vehicle-en-temps-real-per-millorar-el-servei/","referenceExcerpt":"utilitzar la intel·ligència artificial aplicada a les imatges captades per les més de 4.500 càmeres","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}