3D Information Extraction Of Vehicle Based On Monocular Image Sequence

As an important part of video detection technology,3D information extraction of thevehicle has very important function and meaning to the traffic video detection. The mainresearch content of this paper is3D coordinates of vehicles’ angular points on monocularimage sequence.There are four steps for3D coordinates in this paper: from camera calibration,characteristic angular points’ projection speed and vehicle speed detection,3D coordinatesextraction. The main work is as follows:1. Camera calibration. This paper constructs1D and2D mapping tables and realizes thetransformation from image coordinate to world coordinate by monocular camera calibrationwith the perspective projection transformation methods; the camera excursion is detected bythe change features of lane lines; the calibration method can accurately complete the mappingtables through verification of the actual traffic video.2. Speed detection. We extract the feature points by the ratio method, and look for themapping table for the initial distance, which is as a reference point to enumerate a series ofspeed estimate the vehicle’s real-time speed; and then, use the algorithm based oncharacteristics to get the angular points trajectories; finally look up mapping table to getprojection speed.3.3D coordinates extraction. We achieve the height of the vehicle’s feature points by thededuced proportional relation between speeds of the feature points and its spatial height; andthen estimate its two-dimensional coordinates by the deduced triangle relation between itsheight and space position; and finally, achieve good effect by testing and verifying them inreal-time scene and indoor simulation environment.