| In order to ensure people’s safety in the road environment to the maximum extent,the major automobile enterprises are committed to the exploration and research of Advanced Driving Assistant System(ADAS).This article takes the front vehicle detection and ranging in the ADAS as the research goal,and proposes a front vision detection algorithm based on machine vision.The overall algorithm mainly includes 4 steps: image preprocessing,hypothesis generation,Hypothesis verification and Distance measurement.Image preprocessing stage: In order to reduce the amount of calculation of the algorithm,first select the effective area of the image,gray it,compare the three commonly used denoising filters,select the median filter with better effect for filtering,and finally equalize by histogram Enhanced image processing for images with too bright or dark scenes.Hypothesis generation stage: For the case where the fixed threshold method and the single OTSU algorithm are not effective in segmenting the shadow area of the vehicle,it is proposed to use the OTSU algorithm twice to have a good separation,and then based on the shape characteristics of the vehicle shadow,the area threshold and shape Features filter out some non-shaded areas.Then,the method of extracting shadow lines and merging shadow lines can effectively prevent adjacent vehicles from being easily missed,and obtain a preliminary region of interest.Finally,according to the symmetry of the vehicle,the false domain vehicle area existing in the area of interest is filtered to obtain the final area of interest.Hypothesis verification stage: In order to take into account the vehicle’s edge features and texture features,combining the advantages of HOG and LBP features,an extraction method for the fusion of HOG and LBP features was proposed.At the same time,PCA was used to reduce the dimensionality of the fused features.Finally,the features were SVM The classifier is trained and classified to verify the presence of vehicles in the region of interest.Distance measurement: By analyzing the transformation between the coordinate systems and the principle of camera imaging,and then calibrating the internal parameters of the camera,combining the calibration parameters and the coordinate system conversion to obtain the relationship,a monocular ranging method based on the geometric model was used,and Ranging experiment of front vehicle in static and road scene.The detection and distance measurement experiments show that the vehicle visionbased vehicle detection algorithm proposed in this paper can effectively improve the detection rate and reduce the false detection rate,and can basically meet the practical application requirements. |
