Computer Vision Development in Barcelona

Tourism / Urban Visitor Management

Barcelona's visitor management authority (coordinating tourism across the city — managing 12.7 million international visitors annually, working with hotels, attractions, transport, and public safety services) needed computer vision to improve visitor flow management, enhance safety at crowded attractions, and provide data-driven insights for sustainable tourism planning. (1) Crowd safety: peak tourist density at La Rambla, Sagrada Familia surroundings, and Barcelona beach areas regularly exceeded safe thresholds — creating safety risks from crushing, impeding emergency vehicle access, and overwhelming local infrastructure. Current crowd monitoring relied on police officers making subjective density estimates — with no real-time data system to trigger crowd management interventions proactively. (2) Attraction overcrowding: major attractions experienced extreme demand variability — Sagrada Familia visitors concentrated in 3-hour morning and afternoon peaks, Park Guell received 80% of daily visitors between 10am and 2pm, and beaches reached unsafe density on summer weekends. No system distributed visitors across time periods or alternative attractions — resulting in poor visitor experience at peak times and underutilised capacity at off-peak times. (3) Transport pressure: tourist flows created predictable but unmanaged pressure on Barcelona's transport network — metro stations near attractions experienced dangerous platform crowding during peak tourist hours, bus routes through tourist areas ran significantly behind schedule due to boarding delays, and pedestrian congestion on narrow Gothic Quarter streets created safety hazards and resident frustration. (4) Privacy requirements: any visitor monitoring system must comply with AEPD (Spanish Data Protection Agency) requirements — no facial recognition, no tracking of individuals across cameras, and no storage of identifiable imagery. Barcelona's citizens and privacy advocates were particularly sensitive to surveillance technology — requiring transparent, privacy-preserving approaches that were publicly defensible. (5) Resident-tourist balance: Barcelona experienced significant tension between tourism economic benefits and resident quality of life. Data on how tourist flows affected specific neighbourhoods, transport routes, and public spaces was needed to develop evidence-based tourism management policies — but this data did not exist in any systematic form.

We developed and deployed a privacy-preserving visual analytics platform over 12 weeks. (1) Anonymous crowd analytics: privacy-by-design architecture — all processing at the edge (NVIDIA Jetson devices at each camera location), with only aggregate statistics (counts, density maps, flow vectors) transmitted to the central platform. No video or identifiable imagery stored or transmitted. Detection: anonymous body detection (not facial recognition) — counting people in each camera frame and estimating crowd density using calibrated perspective models. Flow analysis: tracking anonymous movement patterns through overlapping camera coverage — understanding flow direction, speed, and congregation points without identifying or tracking individuals. Density mapping: real-time crowd density maps for key tourist areas — colour-coded from green (comfortable) through yellow (busy) to red (exceeding safe thresholds). (2) Attraction flow management: camera-based visitor counting at major attraction entry points — real-time capacity tracking for Sagrada Familia, Park Guell, La Boqueria, and 15 other managed spaces. Predictive models: ML-based forecasting of visitor arrivals based on historical patterns, weather, cruise ship schedules, flight arrivals, and event calendars — predicting crowd levels 2-4 hours ahead. Dynamic recommendations: visitor-facing app and digital signage suggesting optimal visit times, alternative attractions, and less-crowded routes — dynamically adjusting based on current and predicted conditions. (3) Transport integration: real-time tourist flow data integrated with TMB (metro and bus) operations — metro stations receiving advance warning of tourist surges (enabling additional staff deployment and crowd management), bus operations adjusting service frequency on tourist-heavy routes, and pedestrian flow data informing temporary road closures and pedestrianisation decisions in the Gothic Quarter. (4) Safety automation: automated density threshold alerts — when crowd density in any monitored area approached unsafe levels, automatic alerts triggered to police control centre with specific location, current density, and predicted trajectory. Escalation protocols: progressive response — first digital signage directing visitors to less crowded areas, then police deployment for crowd management, then access restriction for severely overcrowded locations. Event monitoring: enhanced monitoring during festivals (La Merce, Sant Joan) and major events (MWC, Sonar) — predictive models adjusted for event-specific crowd patterns. (5) Tourism planning analytics: longitudinal data platform providing evidence for tourism policy — visitor volume by neighbourhood, time, and season. Neighbourhood impact scoring: quantifying tourist pressure on specific residential areas — foot traffic volume, density patterns, and temporal distribution. Policy evaluation: before-and-after analysis of tourism management interventions — measuring whether time-slot booking at Park Guell, Gothic Quarter access management, or beach capacity limits actually reduced overcrowding.

Crowd safety: zero crowd-related safety incidents at monitored locations in 12 months (from 8 incidents requiring emergency response in the prior year). Automated density alerts triggered 340 times — enabling proactive crowd management before unsafe conditions developed. Average time from threshold detection to police response reduced from 25 minutes (reactive) to 4 minutes (automated alert). Attraction management: visitor distribution improved significantly — Sagrada Familia area peak density reduced by 22% through dynamic recommendations directing visitors to off-peak times and alternative attractions. Park Guell visitor distribution spread from 80% in the 10am-2pm peak to 60% — with morning and afternoon shoulder periods absorbing redirected visitors. Visitor satisfaction: average visitor experience rating improved from 3.8/5 to 4.3/5 at managed attractions — less crowding improved the experience even as total visitor numbers remained constant. Transport: metro station overcrowding incidents at tourist-adjacent stations reduced by 35%. Bus schedule adherence on tourist routes improved from 72% to 88%. Pedestrian safety incidents in the Gothic Quarter reduced by 45% through evidence-based temporary pedestrianisation during peak periods. Tourism planning: the city's tourism planning department received, for the first time, evidence-based data on tourist flow patterns across all major areas — enabling informed policy decisions. Neighbourhood impact assessments identified 4 residential areas experiencing unsustainable tourist pressure — leading to targeted management interventions. Privacy: zero privacy complaints in 12 months — the privacy-by-design approach (edge processing, no identifiable data, transparent public communication) satisfied both AEPD requirements and citizen expectations. Annual value: EUR 18M (safety — avoided emergency response costs, reduced liability, and prevented crowd-related incidents), EUR 15M (visitor experience — improved attraction experience driving visitor satisfaction and return visits), EUR 12M (transport efficiency — reduced overcrowding and improved service reliability on tourist routes), EUR 8M (sustainable tourism — evidence-based management protecting residential quality of life and the city's tourism brand), EUR 5M (operational efficiency — automated monitoring replacing manual police crowd estimation and report-based decision-making) = EUR 58M. Development cost: EUR 290,000. Payback: under 2 days.