Research On The Method Of Vehicle Distance Measurement And Warning Based On Monocular Vision

With the increase of the number of vehicles and the increased complexity of road traffic,it is difficult to judge the safe distance in various traffic conditions only by drivers.In order to reduce the driving pressure of drivers on complex roads and improve the safety of drivers and vehicles,studying the method of safe distance warning is very important for society and economy.The main achievements and innovations of this paper are as follows:1.Before using Histogram of oriented grandient to make training set for vehicle recognition,the noise in image is filtered by color space normalization,which improves the speed and accuracy of image processing.In order to accelerate the recognition,a method of image spatial classification is proposed,which improves the speed of vehicle recognition while ensuring the accuracy.In the process of constructing Histogram of oriented grandient training set,YCrCb color space histogram and space block color map are added to the training set,which increases the dimension of training set and improves the accuracy of vehicle recognition.For vehicle recognition,a sliding window method with different schemes for different regions is proposed,which reduces the area of vehicle recognition and improves the speed of model recognition.At the same time,a new filter sideway is proposed to remove false positives and improve the accuracy.2.In Lane recognition,the image is repaired by acquiring the distortion matrix after camera calibration.According to the characteristics of lane in the image,a specific convolution template is designed for threshold segmentation.At the same time,the characteristics of lane information in different color spaces are analyzed.A method of lane fusion in multi-color space is also proposed to improve the preprocessing effect.In the perspective transformation of lane,the bottom zoom strategy is applied,which not only preserves the lane information and visual field,but also eliminates the perspective effect of optical lens.The existing problems of traditional lane detection algorithm are analyzed,and a lane detection algorithm based on virtual sensor is proposed,which solves the problem that lane can not be recognized because of the change of light source and bending lane can not be recognized.3.The relationship between the pixel coordinates and the world coordinates is established,and the advantages and disadvantages of the distance estimation model based on monocular vision are analyzed.According to the selected model,the geometric relationship of the ranging model is obtained.According to the geometric relationship,the improved method of real-time acquisition of camera elevation angle based on parallel lane line and the results of vehicle and lane line recognition are combined to realize the model.4.The typical safety distance and warning model of vehicles is studied,and the acceleration curve of vehicle braking is analyzed.The safety distance warning model is established.The relationship between safety distance and relative speed in different environments is obtained.Based on the previous research results,the safety distance warning system of vehicles is realized.