The Road Recognition Based On Machine Vision Of Unmanned Vehicle

The visual navigation system of unmanned vehicle can identifies the driving region truly based on the visual information. The road recognition algorithm in visual navigation has attracted wide attention of domestic and foreign scholars. Some road recognition algorithm has the problem of poor adaptation and low recognition rate. This paper will study the road recognition algorithm and the pose parameter’s calculation of the vehicle based on machine vision in complex structured road environment, aiming to proposed strong adaptability, high robustness, and good real-time road recognition algorithm. Provide the correct decision for unmanned vehicle’s lateral control.The unmanned vehicle walking on the campus road or the structured road. It will contain rich information in the road image from the camera. In order to identify diversity road information and computes the pose parameters of the unmanned vehicle accurately and quickly, this paper has done the following three research work. Firstly, in the light of over-segmentation problem in the watershed transform, this paper combined with fuzzy reasoning technique, labeled pixels with "fuzzy" in the road information reasonably, and by using the watershed transformation thoughts to divide driving region from the road image information. Secondly, complete the lane markings (or the lane boundary) fitting in the diving region. The linear model of structural road has the problem of poor adaptability, so in this paper, proposed the line fitting approximating method, which divided nearby region of interest in the driving region, using the Hough transform and feature constraints to find the pixel in this region, and completed the least squares line connecting fitting. Finally, combined camera internal and external parameters with characteristic pixel coordinates value in the road boundary, according to the relationship between the world coordinate system and image coordinate system, calculate the pose parameters of unmanned vehicle relative to the road ahead, complete the visual understanding of road ahead.Through a large number of experiments and analysis of data errors make clear that the error of camera calibration algorithm is small. Processing every frame consuming time is far less the ideal time. At the same time, road recognition algorithm has better effect in many complex conditions. The pose parameter of this car is truly. Therefore, the road recognition algorithm which this paper put forward meet each indexes of the evaluation system of the algorithm, makes unmanned vehicle can meet the demand of many different road condition identification, has great practical value.