Home » Technology » Capability

Object Recognition

Deep Vision's data abstraction technology delivers real-time object recognition.

Deep Vision's novel concept of data abstraction quickly transforms abundant sensor data into a form that is easily classified and efficiently analysed.

The abstractions created from the raw data accurately describe the content of the data. By utilising the descriptive properties of the abstractions, couple with their invariant properties (e.g. Rotation, skew and scale) the perception, recognition, and differentiation of objects is enabled.

Deep Vision's data abstraction technology operates with a throughput of 100+ frames per second in complex and cluttered environments and under diverse lighting conditions, both spatial and temporal.
Exploitation Value

  • Target detection, identification, and locking
  • Automatic Target Cueing (ATC)
  • Moving platform, moving target location and velocity estimation.
  • Autonomous systems – sense and avoid, search and destroy, etc.
  • Missile seekers
  • Weapon sub-systems – sighting, fusing etc.

Features

  • Object classification
  • Feature extraction
  • Multiple objects per scene
  • Robust to occlusion
  • Characteristic of Dynamic Template Matching
 Operating Facts

  • Operating System: Any (GNU/Linux recommended)
  • Hardware Requirements: None
  • Sensor Modalities: Visual, Thermal, Sonar
  • Timings†: 100+ FPS
  • Runtime Memory Requirements†: 300 KB
  • Storage Requirements‡: 1.1 KB

† Typical. Based on a 640 x 480 data set
‡ Typical. Based on 45 abstractions (avg. 25 symbols each)

Input Requirements

  • Archived video and images
  • Real-time acquisition from visual, thermal, or sonar sensors.
Examples

1. Recognition of the four playing card suits (hearts, clubs, diamonds, and spades). The suits are hand drawn on a piece of paper to show that the technology is not bound to any particular style. 2. Recognition of characters on a license plate being held in outside in front of a window. It should be noted that the technology is not searching for pre-defined symbols (as per template matching), it is actually recognising the characters. The use of a license plate is arbitrary - it could be a sign on a wall, markings on a ship, name badge on a chest, etc.