In Mobileye’s Vehicle and Pedestrian detection systems, Mobileye uses a classic approach of Model based detection where a “typical set of models” (depending on what is being detected) is employed. These models contain sets of standardised models, for example in pedestrians arms, legs, head and torso. Feature classifiers are then used and the system is trained by a process of statistical learning.
When moving to Any Object detection, the world is a far more complex place, as effectively the desire is to detect any object that may represent a threat to the host vehicle, or may be threatened by the host vehicle. Essentially this can mean any object in the camera’s Field of View. Depending on camera image resolution, ranges will vary, but Mobileye aims to detect general objects initially to a range of 30-40 metres and predominantly in urban scenarios.