In an era where data-driven decision-making defines success, organizations struggle with accurately monitoring crowd flow, optimizing resources, and ensuring security in high-footfall environments. Our AI-enabled People Counter leverages machine vision, advanced algorithms, and edge GPU computing to provide precise occupancy data in real-time. By combining accuracy, scalability, and ease of integration, this solution transforms raw surveillance feeds into actionable insights for industries like retail, transportation, healthcare, education, and hospitality.
Challenges: Designing Proactive and Intelligent Crowd Management:
- Manual monitoring methods are error-prone and resource-intensive.
- Traditional CCTV systems lack actionable analytics.
- Businesses face difficulty in managing peak hours and workforce planning.
- Security lapses due to inaccurate or delayed crowd insights.
- Limited integration with modern IoT/cloud ecosystems.
The Solution — “VisionEdge™ Footfall Intelligence”:
We developed VisionEdge™ Footfall Intelligence, an advanced AI-powered people counting solution that combines custom-trained head detection models, machine vision, and edge-based real-time processing. Built on NVIDIA Jetson hardware, the system delivers lightning-fast, high-accuracy detection without compromising privacy, since it relies on head detection instead of full-body tracking or facial recognition.
The device is designed to operate autonomously at the edge, ensuring minimal latency and reduced cloud dependency, while still maintaining seamless integration with cloud platforms for centralized monitoring, analytics, and reporting. This hybrid approach not only improves efficiency but also ensures scalability across multiple locations with consistent accuracy.
How It Works ?
- Detection: Camera installed at entrances captures FHD (1920×1080) video.
- Inference: Onboard Computer Vision based algorithms detect heads for precise counting.
- Classification: System identifies IN/OUT movement to calculate occupancy.
- Dashboard & Apps: Live counts and historical insights accessible via web & Android apps.
- Cloud Integration: Data transmitted securely to AWS for advanced analytics, if enabled.
Key Takeaways
- High-accuracy footfall detection.
- Real-time monitoring & live dashboards.
- Wired/Wireless connectivity (Gigabit Ethernet, M.2 Wi-Fi).
- Anonymous detection (no personal data captured).
- Easy installation & maintenance.
- Mobile & desktop apps for instant access.
- AWS IoT Core integration for scalable deployments.
Why Choose People Counter™ by Dotcom IoT ?
- Built on NVIDIA Jetson hardware and powered by our proprietary computer vision based algorithms, Counter™ ensures unmatched accuracy, scalability, and reliability.
- From retail stores to airports, it delivers the insights businesses need to optimize resources, enhance security, and boost customer satisfaction.
The VisionEdge™ People Counter is built on an edge-first architecture that combines an FHD camera with the NVIDIA Jetson Nano for real-time AI inference. Incoming video streams are processed locally through deep learning models to detect and count people with high accuracy, before being passed to a lightweight dashboard for instant visualization. For extended monitoring and analytics, the system can seamlessly sync with cloud platforms, enabling centralized reporting, remote access, and continuous model improvements.
Device Layer (VisionEdge™ Counter Unit)
- It has On-Device AI, comes with Custom-trained Computer Vision-based head detection models run directly on the device, ensuring privacy-focused and accurate people counting.
- It comes with a Camera Interface equipped with MIPI-CSI or USB camera modules; optimized for head detection in varying lighting conditions with real-time adjustments.
- The Processor (SoC) is powered by NVIDIA Jetson Nano with Quad-core ARM CPU and GPU acceleration, enabling smooth edge-based AI inference for real-time people counting.
- Its Memory supports 4GB LPDDR4 RAM with microSD-based storage, sufficient for high-speed video processing and local buffering of detection data.
- The Application Layer features Local dashboard interface for live monitoring, device health, and simple configuration (Wi-Fi setup, threshold settings, and calibration).
- The Connectivity Offers Gigabit Ethernet, Wi-Fi, and USB 3.0 for data transfer, integration with cloud platforms, and centralized reporting.
- The Device Runs on 5V/4A power supply with active cooling, ensuring reliable 24/7 continuous operation in commercial environments.
Software & Middleware Layer:
- Image Acquisition Module: Captures raw frames from the USB camera interface, handles frame formatting, color space conversions, and optionally cropping/ROI selection.
- Preprocessing / Enhancement Module: Performs operations like resizing, normalization, brightness/contrast adjustment, and noise filtering to prep frames for inference.
- AI Inference Engine: Runs the head/person detection model using TensorRT / CUDA optimizations on the Maxwell GPU. Counts people entering/exiting frames.
- Counting & Logic Layer: Applies rules and logic to infer IN / OUT events from detections, handles debouncing, boundary thresholds, and occupancy calculations.
- Local Data Service / Buffering: Stores short-term logs locally to prevent data loss in network outages.
- Communication / Sync Module: Handles network connectivity (Ethernet / Wi-Fi), packetizing and sending data (batched or real-time) to cloud / central server if enabled (e.g. via AWS IoT).
- User Interface / Dashboard: Local desktop app and mobile app interface to view live counts, historical data, charts, and configure device parameters (camera, thresholds, sync settings).
Communication & Cloud Layer:
- Cloud Sync (Optional): Using AWS IoT / MQTT device uploads periodic batches or live streaming of counts, occupancy, and alerts.
- Cloud Storage & Analytics: Data ingestion into DB , long-term storage, analytics pipelines, trend modeling, alerts, reporting.
Architecture Flow:
- The Camera captures FHD (1080p) frames through a USB Webcam into the Jetson Nano.
- The Preprocessing module readies data (resizing, normalization) for the Inference Engine, which runs detection models using GPU acceleration.
- Output detections feed into Counting Logic, which interprets true IN/OUT events and computes occupancy.
- Results are logged locally and optionally pushed to the cloud via the communication module.
- Stakeholders view real-time and historical statistics via local UI and/or cloud dashboard.
- Transition to YOLOv11/transformer-based vision models for higher accuracy.
- Heatmap & dwell-time analytics for deeper behavioral insights.
- Integration with smart building automation (lights, HVAC).
- Predictive analytics for peak hour planning and crisis management.
