
Real-time multi-task vision smart agent for fisheye cameras and normal cameras
- Supports RGB and Grayscale cameras.
- Supports front and rear cameras.
- Focused on vehicle ADAS functions and automated road crack detection systems.
Product description

Currently supported end-user-functions
- Bird eye view.
- Objects detection and tracking.
- Semantic segmentation.
Supported systems
Nvidia GPUs (tested on Jetson Nano Maxwell GPU and RTX architectures).
Product facts
System requirements
- RTX 2060 GPU or Orin Nano
- RAM Footprint: 2 GB
- ROM Footprint: 100 MB
Software requirements
- Linux / QNX
- PyTorch 2.1 C++ libraries
- CUDA 11.8 or more
- Available as ROS node.

Future features
- More than 20 different automotive object classes to be detected.
- 3D Bounding box detection and tracking.
- Camera soiling detection and warning.
- Automatic realignment.
- Multi-camera synchronization stitching.
- Real-time 3D HD map generation.
Synthetic training system
- Available for more than 5000 km of driving.
- Based on unreal engine with realistic scene generation.
- Supports OEM custom training scenarios.
- Additional real-world training scenarios planned.
Fisheye ISP & Bird eye View Projection
Image signal processing
- Supports grayscale and RGB cameras.
- Camera automatic calibration algorithm available.
- Frame unwrapping.
Supported KPIs
- 30 fps @ 8 MP
- 60 fps @ 4 MP
Birdeye view projection
- Configurable ROI.
- Camera calibration algorithm available.
Objects Detection & Tracking
Object detection features
- Detects objects classes and 2D-bounding boxes.
- Real-time classification.
Object detection KPIs
- Up to 10 different automotive specific classes, including vehicles, pedestrians, bicycles, lanes, and traffic signs.
- Up to 20 objects concurrently detected at real-time.
- Detection rate @ 10 fps.
Object tracking
- Tracking static and dynamic objects.
- Up to 3 seconds out of view objects tracking.
- Detects objects velocity, position, and direction vectors.
Object tracking KPIs
- Up to 20 2D-bounding boxes, top view 2D map.
- Detection rate @ 10 fps.
Semantic Segmentation
Semantic segmentation & classification
- Detects 10 different automotive specific classes including vehicles, pedestrians, bicycles, lanes, and traffic signs.
- Real-time classification.
- Detects free-space.
- Generates warnings for vulnerable objects.
Semantic segmentation KPIs
- Up to 10 different automotive specific classes, including vehicles, pedestrians, bicycles, lanes, and traffic signs.
- Detection rate @ 10 fps.