Course Content

Driving at night is usually more dangerous than driving during the day. This raises the importance of maximizing a driver’s forward vision by using the vehicle’s high-beam headlight for driving safety purpose. Nevertheless, a recent study by U.S. Department of Transportation shows that, on average, drivers use their high beams less than 25% of the time during which conditions justified their use. This motivates the automobile industry to look into various intelligent headlight control (IHC) systems, to aim for automatically and optimally controlling the headlight of an automobile during a night-time drive. IHC is increasingly be- coming an important aspect of an advanced driver assistance system. With its help, drivers no longer need manually and repeatedly switch between high beam and low beam, and can thus concentrate more on the actual driving.

To our best knowledge, there are not many published research efforts in this area, yet there are a few such systems being prototyped or deployed in the current market. For instance, the Smar tBeam , developed by Gentex, TM uses a customized and forward-facing CMOS image sensor to acquire images in front of the vehicle, which are then processed to detect the existence of headlamps of oncoming traffic or taillamps of preceding vehicles. Appropriate headlamp switching is then performed based on such detection. With similar ideas, Mobileye developed an adaptive headlight control system which also considers the scenario of brightly-lit/urban areas, while Smar tBeamTM does not. Recently, Mercedes-Benz announced that it will integrate an Intelligent Light System and Adaptive Highbeam Assist system into its future models.

Generally speaking, a common approach to intelli headlight control is to detect potential light objects u some image processing algorithms, then apply certain heu tic rules to decide if high beam should be used or not. W such solution is relatively easier and quicker to deve it usually suffers from the drawbacks such as difficult deployment in different geographical regions, lack of rob ness to the change of weather and road conditions, as as expensive system fine-tuning. Consequently, mach learning based approaches become more preferable. such effort is reported in, where a Real-AdaBoost le ing machine is applied to train four classifiers for small/n small headlight, and small/non-small taillight, using var appearance-based features. Nevertheless, other important tures such as motion, which could have helped boost learning performance, are neglected.

Moreover, street lig as another important light object, are not considered. In this work, we aim to develop an intelligent head control system using a forward-facing camera sensor works at real time. Our target is to detect oncoming leading traffic as far away as 1000 meters and 400 me respectively, on flat roads under dry weather conditi When there is an overtaking vehicle, the system sh switch to low beam as soon as possible. Finally, we also detect urban areas that are well-lit with street lig where the drivers should also use low beam. Considering that the system should easily adapt to di ent topography, weather, road signage, and specific veh styles in different countries, we have explored two mach learning based approaches.