| |
| Eyeing the future - development of the monocular sensor.
|
10/11/04 |
|
An Israeli research and development company has produced a single camera optical sensor which, it claims, is able to perform all the functions normally associated with multiple sensor devices.
Using principles of perception and distance borrowed from the human eye, the company has developed a silicon chip capable of the real-time identification of objects and the accurate measurement of distance. It is also developing a series of algorithms capable of identifying vulnerable pedestrians.
"The principle is really quite simple" said Dr Meny Benady, Vice President Marketing and Business Development with Mobileye, a Jerusalem-based company, "and is predicated on how the human eye makes sense of its surroundings. The system has been trained on thousands of images of pedestrians and the resultant characteristics learned by the system. A suite of pattern recognition algorithms is able to match them against what it sees on the road."
At about a fifth of the cost of radar and lidar devices, the monocular EyeQ system-on-a-chip architecture may still find itself struggling to be heard in an increasingly crowded marketplace. Others, including Siemens and the Denso Corporation, are also working along the same lines to fully optimize the inherent potential of the CMOS technology on which the EyeQ system works. The current advantage that Mobileye enjoys is in the use of a single camera and chip, an advantage that may not last for the simpler applications such as lane detection. Indeed, Honda have already installed a single camera system in their Inspire model, currently available only within Japan. It is, as Mobileye were quick to point out, only designed to deal with lane departure warning, a comparatively simple task compared with the objectives set by the Israelis.
"Our choice of a monocular configuration (single camera) is crucial in the context of an automotive high-volume application," said Professor Amnon Shashua of the Hebrew University in Jerusalem and a co-founder of Mobileye. "The conventional way computer-vision technology tackles 3-D scenes is through a stereo set-up, which makes the scene depth-map readily available through stereopsis. The depth-map can be useful for separating the objects of interest, such as vehicles and pedestrians, from the background. The problem is that a stereo set-up capable of measuring distances of over 20 meters requires a significant baseline. This makes the application cumbersome and requires synchronization and fixed calibration between the two cameras to obtain reliable depth measurements. The results may be difficult to guarantee over time in a production vehicle.
In the short term, a joint development agreement signed with the Denso Corporation of Japan in July this year is likely to be of significant help in maintaining the company's present lead, while the extremely small size of the device - about the dimensions of a business card - may add to the attractiveness of the EyeQ for OEMs struggling to find room for the range of new, must-have electronics now becoming available.
"We are looking to position ourselves as a Tier 2 supplier," said Dr Benady. "The precise application will then be a matter for the OEM but our platform is capable of being used to deal with lane departure warning, forward collision warning and headway monitoring and in the near future also with pedestrian recognition, which is a hugely complex task . The products will be in production vehicles in the space of the next 2 to 3 years."
Yet part of the difficulty facing developers of CMOS-based sensors like Mobileye, is the enormous amount of processing power required to deal with the images in the short time available. Optical sensors are not, generally speaking, effective over about 60-100 metres and while better than radar or infrared at classification of an object, are unsuitable for longer range work. The result is that, at the high traffic speeds at which driver assistance systems are required to operate, there is very little time for the sensor to do its work.
Benady believes they have overcome the problem with a single chip processing platform capable of dealing with 30 frames a second - comparable to radar operating at 10 Herz.
"Using one camera instead of two substantially reduces the amount of data needing to be analysed," said Benady, "In addition to this, we have developed the EyeQ, a dedicated chip on which there are six internal processing units, four ASBs (Application Specific Blocks) each specializing its own task in addition to two general purpose ARM CPUs. Of the four ASBs one deals with lane detection and road geometry, a second deals with pattern recognition, a third is used for target tracking and the fourth is concerned with general image processing."
In the continuing trend towards the provision of safety-critical sensors in vehicles, there is a clear need for systems capable of dealing with the highly complex environment provided by the urban scene and in this context optical systems, as opposed to radar or infrared, seem to offer the best hope. Within this sphere questions of reliability, cost and space inevitably raise their heads, questions which Mobileye seem confident of being able to address.
|
Further information, contact:
1. Dr. Meny Benady, VP Marketing and Business Development for Mobileye. Tel: :#972-2-5417305, E-mail meny.benady@mobileye.com
2. ARM re the CPU: www.arm.com |
|
