A detailed look at the sensor fusion pipeline, AI classification architecture, and the engineering decisions that make iSpy reliable in real operational environments.
The TI IWR6843 mmWave module simultaneously transmits at 24 GHz and 60 GHz. The 24 GHz band provides deeper wall penetration through common building materials (drywall, wood, glass). The 60 GHz band provides higher resolution for fine motion detection, including breathing and micro-movements at close range. The Intel RealSense L515 LiDAR captures 23 million depth points per second across a 70° × 55° field of view.
LiDAR data is processed to construct a real-time 3D map of the accessible space. This geometry model provides structural context for radar signal interpretation — walls, doors, furniture, and room boundaries are identified and used to filter radar returns that originate from known static structures rather than potential targets.
Raw radar returns contain significant clutter from static objects, HVAC systems, appliances, and environmental vibration. iSpy's clutter reduction algorithm uses the LiDAR geometry model to mask known static surfaces and applies adaptive filtering to remove periodic interference patterns. This step is critical for reducing false positives in real-world environments.
The cleaned radar signal is passed to a machine learning classifier running on the NVIDIA Jetson edge-AI platform. The classifier is trained on a dataset of human movement signatures across a range of building materials, distances, and movement types. It distinguishes human micro-movements from pets, appliances, and environmental noise, and outputs a confidence score for each detected target.
Each detected target is assigned a confidence level based on the consistency of the signal across multiple radar frequencies, the quality of the LiDAR geometry context, and the classifier's output probability. High-confidence detections represent strong, consistent signals across multiple sensors. Low-confidence detections flag ambiguous signals that may warrant additional caution but should not be treated as confirmed detections.
Processed detection data is transmitted wirelessly to the forearm display. The display renders a simplified tactical minimap showing the operator's position, room geometry, and target dots with confidence indicators. Haptic and audio alerts notify the operator of new high-confidence detections without requiring visual attention.
How raw radar and LiDAR data becomes actionable tactical intelligence
| Module | TI IWR6843 |
| Frequency | 60–64 GHz + 24 GHz |
| Antennas | 4 RX / 3 TX |
| Detection Range | Up to 9m (material dependent) |
| Motion Sensitivity | Breathing-level micro-movement |
| Module | Intel RealSense L515 |
| Depth Points | 23M points/second |
| Range | 0.25m – 9m |
| Field of View | 70° × 55° |
| Output | 3D point cloud + depth map |
| Platform | NVIDIA Jetson (series) |
| Processing | Local edge-AI, no cloud |
| Latency | < 200ms target |
| Classification | Human vs. non-human |
| Output | Confidence score + trajectory |
| Connectivity | Wireless (encrypted) |
| Interface | Tactical minimap |
| Alerts | Haptic + audio |
| Durability | Glove-compatible |
| Power | Integrated battery |
| Form Factor | Chest-mounted rig |
| Enclosure | IP67 target |
| Battery Life | 4–6 hours target |
| Weight | < 800g target |
| Connectivity | Wireless to display |
| Logging | Activation time, operator ID, mode |
| Storage | Encrypted local storage |
| Access Control | Role-based |
| Frequency | FCC / ISED certified components |
| Privacy | No continuous recording |
iSpy is designed to be honest about its limitations. Radar penetration is significantly reduced by reinforced concrete, metal, and dense masonry. When the system detects that it is operating in a high-attenuation environment, it alerts the operator that detection reliability may be reduced.
This degraded-material warning logic is a deliberate design choice. An operator who trusts a false negative in a reinforced concrete structure is more dangerous than an operator who knows the system's limits. iSpy communicates uncertainty rather than hiding it.
Standard residential and commercial construction. Full detection capability.
Reduced range and sensitivity. System alerts operator to reduced reliability.
Significant signal attenuation. System issues degraded-material warning. Results should be treated with caution.
Near-complete signal blockage. System alerts operator that detection is unreliable.
Every iSpy activation event is logged with a tamper-evident record that includes the activation timestamp, operator ID, sensor mode, and duration. This log is stored in encrypted local storage with role-based access control.
The logging architecture is designed to support both internal accountability and external legal review. Agencies can export activation logs for compliance reporting, use-of-force review, and legal proceedings.
Every use event recorded with timestamp, operator ID, and mode.
AES-encrypted local storage. No cloud transmission of operational data.
Log integrity verification to detect unauthorized modification.
Tiered access control for operators, supervisors, and administrators.