Estimate crowd count using density map prediction

Use This When

• Building people counting systems for public spaces, stadiums, or transportation hubs
• Implementing occupancy monitoring for safety compliance or capacity management
• Creating retail analytics that track customer density across store regions
• Analyzing crowd dynamics for event planning or emergency evacuation scenarios

What It Does

• Generates per-pixel density map showing probability of person presence
• Produces PI map (person indicator) and lambda map for multi-scale crowd analysis
• Returns total person count as rounded sum of density map values
• Uses vision-language model with CLIP-based text features for zero-shot counting

Works Best With

• Camera feeds → this component → estimate-density-map-peaks for individual locations
• Density maps → visualize-crowd-count for heatmap overlays on original image
• Integration with alerting systems when count exceeds safety thresholds
• Time-series analysis to track crowd growth and flow patterns

Caveats

• Mean/std normalization parameters must match model training statistics
• Count accuracy degrades in extremely dense crowds where individuals overlap heavily
• Requires GPU for real-time performance on high-resolution video streams
• Density map resolution may not capture fine details in distant or small crowd regions