Algorithm Of Road Detection Based On Computer Vision

Road detection technology is the important perception part of the driverless carsystem and driver assistance system. It provides critical information for drivers or thedriving decision part. The road detection method is often effected by the shadow, ve-hicles and heavy traffic situation. On the non-structured road, the structure infor-mation is always lacked. For the heavy traffic situation and the suburb road, structureroad information is always covered or indistinct. So it is important to develop the al-gorithm which is less independent on the road structure information.The main content in this paper include:First, discus the detection method of horizontal line and vanishing point. For thevanishing point, this paper compare the difference of machine learning algorithm andthe algorithm based on road plane hypothesis and then analyze the effect of road gra-dient to the horizontal line detection. To the detection of the vanishing point, this pa-per propose a new design approach of the texture detected filter and compare differentlocal optimize algorithm. Then propose a new vanishing point estimate approachbased on the EM algorithm. And analysis its performance.On the area of road region recognition, this paper utilize the information fromcolor, texture in the super-pixel, adding some road geometry characters such as hori-zon line, vanishing point and binocular information to detect road area. Then use theregression Adaboost classifier to recognize the road region from the super-pixel. Asthis method has less dependence on road structured information, it could cope withmore complicate situation.Finally, this paper estimate the motion posture of the car by GPS, accelerometerand gyroscope. And build the relationship of the road geometry characters and roadregion between the front and rear frame images based on the motion of the car. Uti-lizing this relationship, this paper map the front image to the rear, combine the infor-mation to enhance the detect result.In the end, this paper introduce a real system build for the road image sampling,also including the sensors, the software and the hardware. Then exhibit some experi-ment about system delay and sensor calibrate. And build a road database based on thissystem.