Research On Mosaic Technology Of Road Capture Image Based On Feature Points

The vehicle violation capture system is one of the important intelligent transportation application devices.The system collects at least three images: the vehicle violation instant image,the license plate number image,and the surrounding environment image of the vehicle.The lane camera has a small shooting range and high resolution for collecting the first two images;the panoramic camera has a wide shooting range and low resolution for collecting the third image.However,this method of forensic can not obtain complete high-resolution vehicle images in the case of over-speed line driving and multiple accidents occupying multiple lanes.For this reason,this paper proposes a feature-based road capture image stitching technique for multiple lane camera acquisition.The images are stitched together.The main research contents and results are as follows:Firstly,this paper proposes a method for screening images of road capture images based on Euclidean distance,which is used to screen synchronous captured image sequences.The method determines the image capturing time by calculating the distance between the same vehicles in the adjacent image sequence,and takes the closest image as a synchronous captured image,and then uses the same as the reference point to filter out the synchronous captured image sequence.Experiments show that the method can screen out the sequence of synchronous captured images from the sequence of road images that are not synchronized.Secondly,this paper proposes a method for evaluating the effectiveness of screening images based on improved cosine similarity.The method calculates the cosine similarity of the non-accurate overlapping area and the non-overlapping area of the image respectively,and performs weighted averaging on the calculated results to reduce the influence of the non-overlapping area on the overall image similarity.Experiments show that the most similar image obtained by this method is consistent with the image selected by the image screening method,that is,the image screening method is effective.Thirdly,this paper proposes a synchronous shooting image registration method based on SURF algorithm.The method uses the SURF algorithm to extract image feature points,uses K-d tree based KNN algorithm to match feature points,and uses RANSAC algorithm to purify matching points.The matching experiment is carried out by using the synchronous shooting image and the equal acquisition serial number image.The experiment shows that the matching accuracy of the synchronous shooting image is higher than that of the equal acquisition serial number image,and the matching correct rate can reach 97.1%.Fourthly,this paper proposes a road capture image fusion method based on improved optimal seam-line.The algorithm enhances the detection gradient in horizontal and vertical directions by improving the traditional Sobel operator,and introduces a weighted average fusion algorithm to smooth the transition splicing crack.Experiments show that the algorithm can avoid the lane line,solve the "ghost" and stitching crack problems,and synthesize high-resolution multi-lane images.The feature-based road capture image stitching technology proposed in this paper can select the synchronous captured image sequence from the asynchronously acquired road image sequence,and the correct rate of registration can reach 97.1%.At the same time,it can solve the "ghost" and stitching cracks.It can also solve the problem that the road violation capture system cannot obtain evidence in the case of overspeed line driving and accidents occupying multiple lanes.