Study On Preceding Vehicles Detection With Muti-features And Distance Measurement Based On Monocular Vision

With the development of social economy, the total amount of vehicles has increased dramatically. Vehicles bring convenience to our daily life, while at the same time they lead to the high level of the accident. The auxiliary system of safe driving, which is an important part of intelligent vehicle, to a certain extent, not only can reduce the pressure of the driver’s driving but also form effective protection to the traffic safety system. Because of the advantages of the vision applied in detection aspects such as its wide detection range, rich information, consistent with human habits of observation, the relatively low cost by the wide attention of the society has attracted extensive attention of the society.Based on the question of the numerous and false target in a single image for vehicle detection for the monocular vision system, a method including lane detection and statistical adaptive gray-level threshold for image segmentation which is get by the mean and variance of gray value of the road, is proposed. And the candidate area of the vehicle is acquired by the shadow line compressed in the binary image; Identification of the vehicles is mainly realized by the muti-feature information of vehicles including texture,shape, symmetry extracted from the candidate areas which are relatively difference from other false target. A single comprehensive threshold eliminating the false target to achieve the purpose of identification is calculated by the Fisher criterion determining the weighted coefficients of the muti-feature. Multiple feature fusion weighted coefficients equal to the vector which consist with the direction of one dimensional linear in the space of multidimensional characteristics. The threshold is get to classify categories by the average and the numbers of a large number of positive and negative samples in statistics on the basis of the experiment many times. Range model is mainly established by the data regression method depending on the relationship between the target position information in the image and the real distance to the preceding vehicle. And the built data model can be used to do an experiment which verifies the accuracy of distance measurement and verification.Algorithm in this paper for different scenarios, road vehicle target detection accuracyhas reached 95.0%, which can eliminate the majority of false objects inside and outside road-region; when data regression model is used to estimate the static vehicle distance, the error of result is within the acceptable range verifying the accuracy of distance measurement.