Our team has developed a prototype of  an ADAS-sensor product, with a set of ADAS algorithms running on high-performance embedded platform. These prototypes are built on a set of ready modules  such as optical sensors, IMU, GPS and processing units. Based on following results it is possible to discuss the main features of this product with possible customers and partners in case of common development on their platforms and with their requirements. Our ADAS team’s main competences are: ADAS algorithms, System engineering, Mature development processes (Scrum, Kanban, Waterfall), CUDA & OpenCL intensive computation, FPGA programming. Our current achievements in Advanced Driver-Assistance System development are:  Patented stereo algorithm, Robust to calibration errors – up to 3 pixels, Target platform – Xilinx 7020, HD resolution, High performance – 20 fps, Low latency – 50 ms, Sparse disparity map –  up to 120000 features.

Main Directions

Collision avoidance  - based on stereo-vision approach and object detection/recognition

AVM – Around view monitor for parking.

ACC – Adaptive cruise control, integration with camera for more reliable behavior in case of not straight-forward movements.

Algorithms integration - building of full chain of algorithms, integration and lab tests.

Dataset gathering – building of our own dataset for testing.

Embedded platform selection/optimization  - possible target platform selection, performance optimization, algorithms porting.

Cameras calibration – optical system calibration for obtain maximum precision and quality.

Prototypes on construction and testing – building

of prototypes together with Topcon for algorithms in-field evaluation.

OUR KEY COMPETENCES 

 ADAS algorithms

 System engineering

 Mature development processes (Scrum, Kanban, Waterfall)

 CUDA & OpenCL intensive computation

 FPGA programming

•Patented stereo algorithm

•Robust to calibration errors – up to 3 pixels

•Target platform – Xilinx 7020, HD resolution

•High performance – 20 fps

•Low latency – 50 ms

•Sparse disparity map –  up to 120000 features