Vehicle Detection

The Mobileye Vehicle Detection (VD) algorithms recognize all motorised vehicles such as cars, motorcycles and trucks, in day and night time conditions (as a stand alone vision system) and was developed and tested on many tens of thousands of hours of driving from all around the world. It accurately estimates the range to the preceding vehicle, comparably to radar based vehicle detection. Vehicle Detection provides the core detection for multiple functions - Forward Collision Warning & pre-crash mitigation, Headway Monitoring & Warning and Adaptive Cruise Control (ACC).

Mobileye's first generation vision based vehicle detection is in production on various Volvo vehicles since 2007 as a radar-vision fusion system and is also included in Mobileye's Aftermarket AWS systems providing Forward Collision Warning and Headway Monitoring & Warning.

Today Mobileye is running multiple series development programs based on its second generation Vehicle Detection (VD2) both as Vision-only applications and enhanced also radar-vision fusion systems for both Warning and Collision Mitigation Systems. These will be launched in vehicles by major global vehicle makers from 2010 onwards, with the launch of the Volvo S60.

Mobileye Vehicle Detection production programs:
Vehicle Detection is combined in all cases with multiple other functions running on a single processor

2007   EyeQ™    Radar-Vision fusion   Volvo S80, V70, XC70 and XC60
2007   EyeQ™    Vision-only               Mobileye aftermarket AWS system
2010   EyeQ2™  Radar-Vision fusion   Volvo S60
2011   EyeQ™    Vision-only               Global auto maker
2011   EyeQ2™  Vision-only               Global auto maker
2012   EyeQ2™  Vision-only               Global auto-maker
2013   EyeQ2™  Vision-only               Global auto-maker

Based on the first generation vehicle detection using EyeQ processor and a VGA resolution imager new vehicles can be detected up to approximately 70 meters and can be tracked to beyond 100 meters in vision-only mode. The system also detects vehicles down to zero speed for Traffic Jam Assist functionality and low speed Forward Collision Warning systems.

The VD2 Vehicle Detection system, based on the EyeQ2 processor, acquires new targets up to 100 meters or more with a VGA resolution imager, and tracks them beyond this. VD2 enhances vehicle detection performance in all aspects, and will enable features such as Adaptive Cruise Control to move to vision based sensing. Future use of automotive qualified high resolution imagers will allow new targets to be detected up to 200 meters. This will allow further enhancement of all functionalities.

The second generation VD2 Vision-only Vehicle Detection system currently detects as follows:
- New vehicles at a distance of 100 meters when using a camera with a Field Of View (Fov) of 38 deg with
  640x480 pixels.
- The detection distance is scalable with the resolution and FoV, so using higher resolution sensors would
  lead to a detection range of up to 200 meters.
- The vehicle detection works in all conditions that a driver can see up to a distance of approximately 100m.
- Built in failsafe mechanisms ensure that the system will become 'non-available' when conditions do not
  allow required performance, for example in cases of heavy fog, extreme weather or camera blockage.
- System availability was tested in multiple production programs and proved to be rated at over 99%, and is
  therefore comparable to other sensing systems such as radar.
- The Mobileye Vision-only vehicle detection provides accurate measurements of range, range rate, lateral
  position & rate that allows vehicle control functions to be derived. Range and range rate measurements
  are comparable to a radar signal, and positioning accuracy is sub-pixel.
- Mobileye vehicle detection utilizes lane detection information and image information to provide a very
  accurate target selection for threat assessment and control, and for collision mitigation by maneuvering.
- Mobileye vehicle detection provides an early detection for vehicle cut-in based on optic flow analysis to
  allow smoother and faster control in all applications.

This figure shows a typical road scene; in which vehicles are detected up to a typical range of 100m. This information is then associated with the results of the LDA module and the vehicles are assigned to lanes. The Closest In Path Vehicle (CIPV) is marked in red. Vehicles in neighbouring lanes are marked in green.