The Ohio State University Corporate Engagement Office

Back to All Technologies

Pedestrian Detection and Tracking with Night Vision

College of Engineering (COE)
Fujimura, Kikuo
Xu, Fengliang
Licensing Manager
Hong, Dongsung

T2003-027 A method to track moving pedestrians and other objects with high collision likelihood using night vision camera on a vehicle

The Need

The precision and clarity of human vision weakens at night, making objects and/or people more difficult to detect while driving. Techniques to develop computer vision that detects humans and other objects include background subtraction, pattern classifiers, and motion patterns. Infrared sensitive cameras are ideal for night time surveillance because humans appear brighter than surrounding objects. Researchers have used these cameras with appearance-based learning, shape-based comparison, and symmetry and histogram methods to detect and locate humans. These multi-step methods use criteria, such as location, size, and road information, to locate pedestrians. However, these methods fail to locate all pedestrians and other objects in the vehicles path. A new method is needed to create an accurate pedestrian detection system that can operate within a moving vehicle to protect both the driver and pedestrians from accidental collisions at night.

The Technology

Researchers at The Ohio State University, led by Dr. Kikuo Fujimura, used night vision cameras to develop a pedestrian detection method that differentiates the movement of objects from the background. Considered complementary to shape-based approaches, the method analyzes night vision video data to differentiate humans motions from the background. The two techniques introduced are a two-stage method for stereo correspondence and motion detection without explicit ego-motion calculation. This method has been verified with experimental results and analysis in comparison to frame-by-frame based pattern-recognition approaches.

Commercial Applications

  • Automotive safety systems
  • Military surveillance
  • Security systems


  • Over 90% of pedestrians can be detected
  • Capable of detecting both people an moving objects
  • Bypasses ego-motion computations to reduce the number of analysis steps
  • Increased detection speed over pattern-recognition methods
  • Suitable for a moving vehicle.