Skip to main content

Occupancy Tracking and Vision-Based Security

While the NDP120 is an acoustic powerhouse, the Syntiant® NDP200 series extends the eFabric™ architecture into the visual domain. Vision-based security at the edge focuses on "Intelligent Gating"—using low-power imaging to perform complex spatial tasks like human presence detection and face recognition without the privacy risks or thermal overhead associated with traditional cloud-based video processing.

  • Face Recognition and People Counting: The NDP200 can be trained to recognize specific faces or count individuals in a frame. This is achieved by extracting key facial landmarks locally. By calculating the Euclidean Distance (d) between a captured face vector and a stored "Golden Template," the system can grant or deny access near-instantaneously.

Formula: Face Recognition Distance (dd)

d=∑i=1n(vcaptured,i−vtemplate,i)2d = \sqrt{\sum_{i=1}^{n} \left(v_{\text{captured},i} - v_{\text{template},i}\right)^2}

(where v represents the feature vectors extracted by the neural network.)

  • Occupancy Tracking and Human Presence Detection: Traditional PIR (Passive Infrared) sensors often fail if a person remains still. The NDP-based vision system, however, uses "Motion-Plus" logic. It detects the specific silhouette of a human being, allowing for accurate occupancy tracking in offices or homes. This ensures that HVAC and lighting systems are optimized based on real presence, not just initial movement.

  • Smart Surveillance and Triggered Recording: The NDP acts as a "Visual Sentinel." It remains in a microwatt state, analyzing low-resolution frames for specific events (e.g., a person entering a restricted zone). Only when the Detection Probability (PdetP_{\text{det}}) exceeds a set threshold does it trigger the high-power system to begin 4K recording or alert security personnel.

Formula: Object Detection Confidence (PdetP_{\text{det}})

Pdet=Area of OverlapArea of UnionP_{\text{det}} = \frac{\text{Area of Overlap}}{\text{Area of Union}}

(also known as Intersection over Union (IoU), used to validate that the bounding box correctly identifies the human subject.)

⚠️💡 The "Metadata-Only" Stream

"To maximize security and privacy, configure your system to transmit only Metadata (e.g., 'Person Detected', 'Count: 3') rather than the actual video stream. This allows the system to remain functional and secure even over low-bandwidth connections like LoRaWAN or Zigbee, while ensuring no private images ever leave the device."