A LANE MARKER DETECTION ALGORITHM SUITABLE FOR EMBEDDED REAL-TIME SYSTEMS

摘要

We present the design of a lane marker detection algorithm suitable for embedded real-time systems. The algorithm was particulary designed for fast execution. The achieved execution time was about 11 ms for a search window of 200×90 pixels on a single state-of-the-art Digital Signal Processor (DSP). The algorithm was integrated into an embedded real-time stereo vision sensor. This sensor was part of the sensor suite of the autonomous vehicle RED RASCAL of the team SciAutonics / Auburn Engineering. RED RASCAL participated at the National Qualification Event (NQE) in the DARPA Urban Challenge 2007 in Victorville, CA, USA. The embedded stereo vision sensor did a great job and detected lane markers and lane borders with a high reliability.