How video perception works
Video perception, also known as computer vision, enables video-based detection and classification of the vehicle’s surroundings, which in turn allows driver assistance and automated driving functions to be implemented in accordance with SAE Levels 0–4 for driving and parking applications. This includes safety functions such as automated braking while driving forward and reversing, convenience features such as adaptive cruise control, active lane change, parking space detection, and parking assist, as well as automated driving on highways, country roads, and in the city.
The software fuses raw video data from one or more camera sensors. It processes image data from the cameras and classifies objects, such as pedestrians, cyclists, and other vehicles, as well as semantic and topographical information about the surroundings, such as the shoulder of the road. To detect and classify objects, the software uses artificial intelligence methods including deep neural networks and accesses global databases when training these networks in order to meet specific local requirements.
More and more, software is becoming a determining factor in the development of automotive electronics systems. It plays a central role, especially in driver assistance. Drivers expect their vehicles to be capable of updates, just like they are used to from their smartphones. In the future, features will be downloaded to the vehicle like apps without the hardware having to be replaced. Video perception from Bosch is a purely software-based solution and can be used on different preselected SoCs (systems on chip).
Along with the camera heads and the ADAS integration platform – the high-performance computer for the ADAS domain from Bosch – customers can put together modular, scalable solutions, from the individual components right up to their complete, customized system.