Research On Assistant Driving Based On Image Recognition

With the progress of science and technology,the economy is developing and the number of vehicles is also increasing.At the same time,with the related technologies are beginning to mature,more and more attention has been paid to the research of vehicular intelligent devices.In many practical applications of vehicle auxiliary technology,lane detection and vehicle detection are the most common and the most tedious problem.Because of the complexity of vehicles,such as different colors or different models,these factors increase the difficulty of vehicle detection.Meanwhile,different external environments or different angles of the device would bring some influence on the final detection results.So,this article has carried on the research to this aspect.The detailed process is as follows:Section 1,Lane Detection.An improved lane detection algorithm based on Hough transformation is implemented.First,the region of interest is selected after image pre-processing.Then the inverse perspective transformation is carried out.Conversion of image angle is easy to detect lane line.After that,using Hough transform and Kalman filter to detect lane lines.Removing the wrong results and outputing the final result.The results proved that the algorithm has good stability and accuracy.There is another algorithm based on color space and edge detection,which could detect lane very vell.Using Lab color spaces in image and counting the distribution of pixels of the binary image to find where the lane is.Then fitting the points to lines.Experimental results show that the lane detection algorithm has high accuracy and adaptability,and it can adapt to most of the curved lane detection.Section 2,Vehicle Detection.A vehicle detection method based on gradient direction histogram feature and support vector machine is implemented.The first step is collecting the required data samples.Then extracting the HOG features from the samples to train the SVM classifier.Before extracting HOG features,changing the RGB color space to YCrCb color space,which could analyze the most suitable channel for extracting features.Then support vector machine learning through data samples to find the feature vectors in data samples,and generating the optimal hyperplane.After that,searching the vehicle targets through the scanning window.