Mobileye® utilizes the unique pedestrian detection technologybased on a mono-camera with EyeQ2 microprocessor. The world-recognized technology is built in vehicles during their manufacture worldwide.
Approach of Mobileye® to pedestrian detection is based on the use of monocamera, and achievements in the image and qualifier detection, with the image processing and flow optical analysis.
First Mobileye’s product for pedestrian detection was introduced in 2009 and installed in a great number of vehicles operating EyeQ2mono-camera and ensuring comprehensive identification of pedestrians in the distance of 15 m. If a pedestrian was detected in the vehicle’s path, the driver would receive audio and video alerts.
At the end of 2009 Mobileye® added a new feature pedestrian collision warning in the Mobileye С2-270 product line.
In the mid of 2010 Mobileye® was the first who implemented on Volvo S60 and V60 the emergency braking innovative solution to mitigate severity of the vehicle collision with pedestrian. Mobileye® vision sensor unit is a key technology for pedestrian detection.
a Picture size: The images of distant pedestrians appear as very small ones. The transverse size is even less.
b High dynamic: The delays in detection should be small, and solutions must be received during several frames.
c Difficult conditions: As a rule, pedestrians are detected in urban environment with a great number of background textures.
d Parametres: The image of pedestrian has no robust edges with a high variation of appearance, and is difficult to trace.
The difficulties with determination of the image sizes are of technical nature Mobileye® found that the production programs will always tend to advancing additional requirements to the vision sensors.
Many efforts have been used to find solutions for correct classification of images of very small size.
High dynamic and difficult conditions for the objects identification require the classification of high accuracy. Intensive researches and search for solutions have resulted in development of the pattern classifier.
Local peculiarities of the object classification are taken from the image depending on intensity and derivatives calculated at one level of pixels or from small areas of the image.
Global features of the image, which reflect links of the images, are integrated in the classification process. For example, long lines of the image that pass through the detection area, issue a negative signal of detection.
An early detection of people, who are in motion is connected with a problem of high dynamics. In their solutions Mobileye® utilize optical analysis of the flow to make out edges of the moving objects.
Today Mobileye’s pedestrian detection feature is active in daytime only. Mobileye conducts further researches and developments to create a technology of NIR filtration of images that will enable this feature to be operated at night hours as well.