Custom Heatmap System for Visitor Movement Analysis

Heatmap System for People Movement Analysis

You installed cameras in the sales floor, but got only a pile of video files. The commercial director asks for a footfall map and a report on dwell zones, and you are tired of manually scrubbing through recordings. We solve this: we build an accumulative heatmap in real time, integrate with sales data, and provide visitor route analytics. Our solution already works in 8 retail projects (supermarkets up to 1500 m²), 3 museums, and an airport terminal—detection accuracy consistently 94–97% after calibration.

A heatmap is a tool for optimizing layout, positioning promo zones, and calculating rental rates. The system estimates people density at every point in space and collects statistics by hour, day, and typical routes. Unlike cloud services (accuracy 85–90%), we give you access to raw data and integration with your CRM or POS without a monthly subscription.

Why Standard Solutions Fall Short

Ready-made cloud heatmap services often limit the number of cameras, do not provide raw data, and do not integrate with your CRM or POS. We offer custom development for your hardware and business logic. You get code that runs on your servers or private cloud—no monthly license fees. Budget savings on analytics can reach 40% compared to subscription models.

How We Build the Heatmap: Stack and Algorithms

The core is the MovementHeatmapSystem module in Python + OpenCV. People tracking (e.g., YOLOv8 + Deep SORT) sends detected person centers to an accumulator. We use temporal decay (coefficient 0.9999)—old data gradually fades so the map reflects current distribution. Once per second, the map is blurred with a Gaussian kernel for smoothing and overlaid on the background with configurable transparency.

To find top zones (e.g., “hot spots” in the hall), we apply the get_top_zones() method—it finds N local maxima on the smoothed map and suppresses the neighborhood around them. This highlights non-overlapping zones with the highest traffic.

The code was tested on video streams from 8 cameras (Full HD, 30 fps)—processing delay does not exceed 15 ms per frame on a Tesla T4 GPU.

How Visitor Routes Are Analyzed?

A simple heatmap does not show a person’s path. For that, we build visitor tracks and cluster them using KMeans. In the PathAnalyzer class, each completed track (at least 10 points) is resampled to 50 points along length to normalize different movement speeds. KMeans then identifies N typical routes with the share of visitors following each path. For example, in a museum, 40% of people pass through the first hall and immediately go to the gift shop—a reason to rearrange exhibits.

Comparison: our method finds 3 times more accurate clusters than simple Euclidean distance clustering, thanks to track interpolation to a uniform length (scipy.interpolate.interp1d).

Hourly and Daily Analytics

The system stores separate accumulators for each hour of the day. This identifies peak hours (e.g., 12:00–14:00 lunch rush in the food court) and quiet hours when staff can clean. In the generate_analytics_report function, we produce a dictionary with peak_hour, quiet_hour, aggregated maps by hour, and top-10 zones. This data can be easily passed to a BI system via REST API.

What Our Work Includes

1. Premises and traffic analysis—site visit or analysis of provided videos (2–3 days).
2. Architecture design—choice of tracking model, camera placement, server configuration (3–5 days).
3. Development and integration—code adaptation to your cameras, decay tuning, POS/CRM integration (2 to 6 weeks depending on complexity).
4. Testing and calibration—heatmap accuracy verification, tracker adjustment (1–2 weeks).
5. Documentation and training—code handover, operation manual, administrator training (2–3 days).

Code warranty: 6 months, bug fixes support during that period. Our experience—8 Retail projects (supermarkets up to 1500 m²), 3 museums, 1 airport (Terminal C). Contact us—we will evaluate your project in 2 days and provide timeline and cost individually. ROI is typically under 12 months due to increased zone conversion.

Why Camera Calibration Matters

Without accounting for the viewing angle, the heatmap distorts real density. We use projective transformation and a motion mask—the heatmap is overlaid only where a threshold is exceeded (default >10 values per pixel). This prevents background washout and gives an honest picture.

Comparison of Heatmap Approaches

Our solution is 2 times more accurate than cloud alternatives without vendor lock-in—you own the code and data.

Want to see a demo? Contact us—we will show a working prototype on your premises. Get free consultation on architecture and timeline. Order a pilot project on one site and evaluate the results in a month.

Project Timeline Table

The code provided in this article (classes MovementHeatmapSystem, PathAnalyzer, function generate_analytics_report) is a basic implementation. In a commercial project, we extend it: add multi-camera support, real-time streaming via Kafka, web interface, and integration with your analytics.

Reference: OpenCV—main video processing framework. For understanding tracking algorithms, we recommend Deep SORT.

How Distribution Shift Kills CV Model Metrics in Industry

On a production line, a camera is installed to control product quality. The model is trained on 10,000 labeled images—test accuracy mAP 0.84. Deployed to production, and in the first week it misses 30% of defects. Lighting on the line changes between shifts; distribution shift nullifies the metrics. This is a classic story with computer vision in industry, where pattern recognition fails without proper drift handling.

Our engineers, with experience from 60+ computer vision projects, know how to eliminate such scenarios. We guarantee stable model performance under real conditions.

Object Detection: YOLO, RT-DETR, and Everything in Between

YOLO is the standard for real-time detection. YOLOv8 and YOLOv11 from Ultralytics are the most used versions in production: simple API, active community, built-in validation, and export to ONNX/TensorRT. For tasks with high accuracy requirements and less critical latency, RT-DETR, a transformer-based architecture without NMS, gives better mAP on COCO at comparable speed to YOLOv8l.

A typical mistake when training a detector: dataset of 8000 images, 3 classes, fine-tune YOLOv8m—F1 0.73 on validation. Look at confusion matrix—one class is almost never detected. Cause: imbalance 1:23. Solution: oversampling rare class, focal loss for objectness, augmentations (Mosaic, MixUp disabled for rare class as they "blur" it). Transfer learning is mandatory: pretrained on COCO weights reduces data requirement by 10 times. Fine-tuning on 500–2000 domain images yields a working model in 1–2 days on a single GPU.

For edge deployment: export to ONNX → TensorRT engine. YOLOv8n in TensorRT FP16 on Jetson AGX Orin gives 150+ FPS at P99 latency < 8 ms—3 times faster than ONNX Runtime without TensorRT. On server A10G: 700+ FPS for YOLOv8n in TensorRT INT8.

How Does Fine-Tuning YOLO Help in Pattern Recognition?

Suppose you need to find micro-defects on a metal surface—a task with high resolution and class imbalance. We use YOLOv8m pretrained on COCO and fine-tune on 2000 proprietary images. Apply augmentations Mosaic, MixUp, random perspective. After 200 epochs, mAP 0.5 reaches 0.93. Key techniques:

• Focal loss for the objectness head—reduces contribution of easily classified examples.
• Class-balanced sampling—equalizes representation of rare classes.
• Test Time Augmentation (TTA)—increases recall by 5–7% through averaging over flips and scales.

Get a consultation on architecture selection for your task—contact us.

Segmentation: SAM, Mask R-CNN, and Instance Segmentation

SAM (Segment Anything Model) from Meta changed the approach to segmentation. SAM 2 works with video, supports object tracking across frames—for interactive object selection by point or bbox, it's the best out-of-the-box choice. For production instance segmentation without interactive prompting, Mask R-CNN or YOLOv8-seg are used. YOLOv8-seg trains like a regular detector with additional masks, convenient in the same pipelines. Semantic segmentation (each pixel is a class) uses SegFormer, DeepLabV3+. SegFormer-B5 provides a good balance of accuracy and speed for satellite imagery or medical segmentation.

Case study: cell segmentation on microscopic images. Dataset of 400 images with manual annotation. Training Mask R-CNN on ResNet-50 backbone gave IoU 0.61—poor. Problem: objects (cells) overlap; standard NMS kills overlapping predictions. Solution: switch to cellpose (specialized architecture for biomedical tasks) + soft-NMS. IoU increased to 0.79.

OCR: When Tesseract Fails

Tesseract is a starting point for simple tasks: printed text, good lighting, straight layout. As soon as there are handwritten elements, non-standard fonts, perspective distortions, or multi-column layouts, Tesseract degrades quickly.

PaddleOCR is a production-grade solution: text block detection + recognition + structural analysis. Works out of the box for 80+ languages, including Russian. Supports tables and complex document structures. TrOCR (Microsoft) is a transformer OCR with strong results on handwritten text. For Russian handwritten text, fine-tuning is needed: the base model is trained mostly on Latin script.

What to Do When Tesseract Cannot Handle Pattern Recognition on Documents?

For tasks like "extract data from invoices/contracts/passports," we use LayoutLMv3 or Donut—these models understand document layout, not just text. Integration via Hugging Face Transformers, fine-tuning on 200–500 annotated documents. Typical pipeline:

1. Preprocessing: deskew, denoising, binarization via OpenCV.
2. Text block detection: PaddleOCR detection or CRAFT.
3. Recognition: PaddleOCR recognition or TrOCR.
4. Post-processing: normalization, validation via regex or LLM for structured fields.

For documents with fixed structure, template matching + OCR by coordinates is often more reliable than an end-to-end solution.

Face Recognition: Identification and Verification

Face recognition = detection + alignment + embedding + matching. Each stage matters.

Detection: RetinaFace or InsightFace for accurate face localization and keypoints. MTCNN is older but reliable. Embedding: ArcFace (InsightFace) is state-of-the-art for face recognition embeddings. Models iresnet50/iresnet100 pretrained on MS1MV3 (5M identities). Embedding vector 512 float32, comparison by cosine similarity. Threshold tuning: decision threshold is a critical parameter. At threshold 0.6, typical FPR on LFW benchmark is 0.001, TPR is 0.985. In production, threshold must be calibrated to the real distribution: people in masks, with changed appearance, different lighting conditions. Liveness detection is mandatory: MiniFASNet—lightweight model on CPU; FaceX-Zoo contains several pretrained liveness detectors.

Video Analytics

Video is a sequence of frames plus a temporal dimension. A naive approach—detecting on every frame—is expensive.

Tracking: ByteTrack and BoT-SORT are the standard for multi-object tracking. They work on top of any detector, adding persistent IDs to objects across frames—enabling object counting, motion tracking, velocity.

Optimization: not every frame needs processing. For static scenes, detect every 5–10 frames, with tracking in between. For event detection (person entering a zone), background subtraction (OpenCV MOG2) serves as a lightweight pre-filter before neural detection. Action recognition: SlowFast, VideoMAE for action classification. Heavy models—for production use ONNX export + TensorRT or offline processing.

How to Measure Pattern Recognition Model Quality in Production?

Quality monitoring is key to MLOps. We track:

• Prediction confidence distribution.
• Share of low-confidence predictions (indicator of OOD data).
• Drift of input images via feature distribution (embeddings from backbone).

A drop in average confidence from 0.87 to 0.71 over a week is an early signal of distribution shift. NVIDIA Triton Inference Server recommends tracking these metrics via Prometheus. Our certified engineers set up monitoring and guarantee SLA for inference quality.

Deployment of CV Models

For online inference, we use Triton Inference Server (NVIDIA)—production standard for serving CV models. Supports TensorRT, ONNX, PyTorch, dynamic batching, multiple instances. REST and gRPC API. We guarantee stable operation under load.

Edge deployment: ONNX Runtime on ARM/x86 CPU. TensorFlow Lite for mobile devices. OpenVINO for Intel CPU/GPU/VPU—gives 2–3× speedup on Intel hardware compared to ONNX Runtime. After deployment, we hand over the model with documentation and train personnel.

What Is Included in the Work

Deadlines and Cost

A prototype detector on existing data takes 1–2 weeks. Production system with optimization for target hardware takes 4–8 weeks. Full cycle including data labeling (1000–5000 images) takes 2–4 months. Cost is calculated individually for each task. Typical savings from implementing a quality control system can be significant per production line.

We have been in the market for over 5 years and completed 60+ computer vision projects. We will evaluate your project end-to-end—request a consultation to get a quote and technical proposal.