Case Study: People Counting System for Retail Spaces

This case study presents the implementation of a computer vision-based people counting system for a retail store chain. The system allows for precise analysis of visitor flows, optimization of operations, and improvement of customer experience through collected data.

Context and Challenges

Client Profile

A national chain of 85 medium-sized clothing stores (400-800 m²) located in shopping malls and city centers.

Initial Issues

• Difficulty in precisely measuring foot traffic and conversion rates
• Staff often poorly distributed relative to peak traffic
• Inability to analyze customer journeys and hot/cold zones
• Need to objectively evaluate the impact of marketing actions
• Obsolete existing system based on unreliable infrared sensors

Business Objectives

1. Obtain accurate data on foot traffic (>98% accuracy)
2. Optimize human resource allocation according to traffic
3. Analyze visitor behaviors to improve merchandising
4. Objectively measure the performance of different stores
5. Comply with GDPR regulations on personal data protection

Implemented Solution

Hardware Infrastructure

• 3-8 cameras per store depending on size (main entrances + strategic areas)
• Discreet IP cameras with wide viewing angle (120°)
• Edge computing on mini-PC for local image processing
• Secure connectivity to the cloud for data aggregation and analysis
• Optional in-store displays for real-time capacity display

Software Architecture

• Techsolut platform for centralized management
• Specifically trained counting and tracking models
• Custom dashboard accessible via browser and mobile application
• API for integration with existing systems (ERP, CRM)
• Secure cloud infrastructure for long-term storage and analysis

Methodological Approach

Phase 1: Needs Analysis and Design

1. Detailed audit of 3 representative stores
2. Mapping of flows and critical measurement points
3. Architecture design adapted to each store typology
4. Definition of KPIs and expected reports

Phase 2: Pilot Deployment

1. Installation in 5 representative stores
2. Fine calibration of algorithms and detection thresholds
3. Training of store and headquarters teams
4. 6-week evaluation period with iterative adjustments

Phase 3: Large-Scale Deployment

1. Progressive installation in the remaining 80 stores (10 per week)
2. Integration with existing enterprise systems
3. Implementation of dedicated support and comprehensive documentation
4. Quality monitoring and regular performance audits

Key Features

Bidirectional Counting

• Precise measurement of entries and exits at each access point
• Distinction of employees via specific badges or exclusion zones
• Management of groups and simultaneous entries
• Error compensation through self-correction algorithms

Flow Analysis

• Heat mapping of movements
• Measurement of dwell times by zone
• Analysis of typical paths and anomaly detection
• Anonymized tracking to respect privacy

Advanced Analytics

• Traffic forecasting based on history and external factors
• Correlation with weather data, events, and marketing actions
• Normalized inter-store benchmarking
• Automatic calculation of key KPIs (conversion rate, cost per visitor, etc.)

Alerts and Automations

• Real-time notification of capacity exceedances
• Security alerts for abnormal situations
• Automatic counting trigger for specific campaigns
• Integration with ventilation/air conditioning systems for energy optimization

Results and Impact

Technical Performance

• Counting Accuracy: 99.2% (validated by manual counting)
• System Availability: 99.8% over 12 months
• Latency Time: < 1 second for real-time data
• Heatmap Accuracy: Spatial resolution of 50cm

Measured Benefits

• HR Optimization: 12% reduction in personnel costs at equal traffic
• Increase in Conversion Rate: +8% thanks to better resource allocation
• Reduction in Checkout Wait Times: -37% during peak hours
• ROI Achieved in: 9 months for the entire deployment

Operational Improvements

• Team planning aligned with actual traffic peaks
• Store reorganization based on flow data
• Objective evaluation of store and team performance
• Reliable traffic forecasts enabling better inventory management

Difficulties and Solutions Provided

Technical Challenges

Variable Lighting Conditions

• Problem: Degraded performance in low light or backlit conditions
• Solution: High-sensitivity cameras and algorithms adapted to light variations

Varied Store Configurations

• Problem: Impossibility of having a single solution for all stores
• Solution: Standardized deployment models but adaptable to each typology

Privacy and GDPR

• Problem: Need to count without identifying people
• Solution: Edge computing processing, immediate anonymization, and no image storage

Organizational Challenges

Resistance to Change

• Problem: Team fears about a "surveillance" tool
• Solution: Transparent communication about objectives, training, and user involvement

Integration with Existing Processes

• Problem: Multitude of systems in place depending on the store
• Solution: Flexible API and standard connectors for different environments

Team Adoption

• Problem: Initial underutilization of available data
• Solution: Creation of custom user profiles and targeted training

Lessons Learned and Best Practices

Key Success Factors

1. Co-construction with the business: Involve operational staff from design
2. Progressive Deployment: Allow adjustment without disrupting activity
3. In-depth Training: Ensure teams know how to exploit the data
4. Change Management: Communicate the benefits for everyone

Recommendations for Similar Projects

1. Clearly identify critical counting points before deployment
2. Plan for a sufficient calibration phase for each site
3. Establish clear governance of collected data
4. Plan regular data exploitation reviews

Evolutions and Perspective

The client now envisions:

1. Feature Extension:
2. Anonymized demographic analysis (age groups, gender)
3. Detection of interest/disinterest behaviors towards products

4. Integration of complementary IoT sensors (temperature, air quality)

5. System Expansion:

6. Deployment in new stores
7. Addition of queue analysis features
8. Integration with the e-commerce system for a unified customer view

Conclusion

This deployment illustrates how a modern computer vision-based people counting system goes far beyond simple counting to become a true retail activity management tool. The progressive approach and emphasis on user adoption were decisive in transforming a technical solution into a concrete operational advantage.

The combination of adapted hardware, performing algorithms, and careful integration with existing processes allowed for rapid ROI generation while laying the groundwork for more sophisticated use of traffic data.