With pure artificial intelligence without computer vision and software, no application can be built. Artificial intelligence has its limitations, e.g. objects that have never been trained cannot be recognized. Therefore the combination with computer vision is necessary. The EYYES software enables the implementation of applications for image recognition and processing.
The Video Software Pipeline from EYYES is a complete software package for processing and evaluating a video signal from the sensor to the monitor. It is the core of all video-based AI use cases. The software can be run on standard hardware or on FPGAs with the appropriate resources as embedded software in products.
The EYYES Video Software Pipeline in detail
Video Input Core
The Video Input Software takes care of the complete part of the camera control as well as the preparation of the video for the image processing.
After the Input Core, the video stream is split. The images flow through the DNN Core Package (Deep Neural Network) for evaluation and classification of the objects. The IP Core is EYYES’ own software processor.
The EYYES Net is a specially trained Deep Learning Network that is trained to recognize all objects on the road. Likewise, our neronal networks can also be trained on all other use cases such as industry or security surveillance.
Video Output Core
With the evaluation of the objects from the DNN Core, the original video stream is supplemented with the marking of the objects with corresponding bounding boxes, so that the recognized objects are visualized accordingly in the video. The image is thus prepared for processing and display on the monitor.
The Box Tracking Library forms the basis for the use cases after object detection. Here we calculate the movement behavior of the objects. This results in possible collisions or accident risks of which the driver and the road users can be warned. This is the basis for all driving assistance functions of a vehicle.
AISoftware pipelineComplete software pipeline for AI products
EmbeddedAIEmbedded AI in FPGA with high efficiency
EfficientIP CoreIP Core with extremely efficient deep learning processing
AccelerationUp to 40 timesUp to 40 times faster neural networks for high performance operation in FPGA.
Lowerpower consumptionUp to 90% lower power consumption due to Deep Learning Accelaration of the network.
SubsequenttrainingThe neural network can be re-trained for all use case requirements.
Functional Safetyfor ISO 26262Functional Safety compliant development for the fulfillment of ISO 26262