RETINA/SENSE

Where this sensor wins

Event cameras are not a replacement for photography — they are a replacement for watching. Anywhere a machine stares at a scene waiting for something to move, the retina approach costs less and reacts faster.

1. High-speed robotics & drone control

  • • Vision-based state estimation for fast flight
  • • Deep learning for obstacle avoidance (learning high-speed flight in the wild)
  • • Trajectory planning through narrow spaces
  • • Perception-aware model predictive control (PAMPC)
  • • Sim-to-real transfer for agile flight

2. Visual SLAM & navigation

  • • ORB-SLAM style camera-based mapping
  • • Feature matching between frames
  • • Loop closure — recognising revisited places
  • • Fusing camera with motion sensors (IMU)
  • • Removing moving objects from the map

3. Driver monitoring — face, eye and blink tracking

  • • Facial landmark detection with Dlib or MediaPipe models
  • • Eye Aspect Ratio (EAR) for blink detection from landmark geometry
  • • CNN-based drowsiness classification on eye-region images

4. Autonomous driving — pedestrian detection

  • • Deep CNN detectors: Faster R-CNN, YOLO and SSD for real-time localisation
  • • Multi-sensor fusion of camera, LiDAR and radar for low-light and occlusion
  • • Speed-accuracy trade-offs with lightweight models on in-vehicle edge hardware

5. Object detection & recognition

  • • Driving: pedestrians, vehicles and traffic signs in real time
  • • Medical imaging: tumours and lesions in CT, MRI and X-ray
  • • Video surveillance: intrusion detection, crowd monitoring, activity tracking

6. Space situational awareness

  • • Telescope-mounted event sensors tracking satellites and debris
  • • Objects detected from LEO all the way to GEO orbits
  • • Works in daylight as well as at night — very hard for conventional CCDs
  • • Low data rate, low power and high speed suit space hardware constraints

Measured results

< 5 ms

end-to-end latency for event-based drone navigation (IoU 0.78–0.83), with up to 20% less flight time than depth-only pipelines

Sanyal et al., 2025

AUC ≈ 0.79

object-tracking accuracy for event-based traffic monitoring using a 0.1M-parameter CNN, inside a sub-1 mW power budget suitable for IoT

Mohan et al., 2020

5,000 fps-equivalent

latency from fusing a 20 fps RGB camera with an event stream — using only the bandwidth of a 45 fps camera

Gehrig & Scaramuzza, Nature, 2024

Future prospects

Autonomous vehicles

High temporal resolution catches fast-moving objects and sudden traffic changes, handles extreme lighting shifts, and fuses with conventional cameras for reliable perception.

UAVs (drones)

Indoor navigation where GNSS is unreliable, motion-blur resistance for rapid manoeuvres, and real-time pose estimation when paired with IMU and visual-inertial odometry.

Smart surveillance

Detects rapid movement, accidents and suspicious activity in poor lighting — well suited to public spaces, transport hubs, hospitals and smart cities.

Try it on your own scene — the lab measures the savings live.

Open the live labView the slides