Study On Measure Distance And Calibration In Video Speed Measurement System

Following with the developing of computer vision technique, using camera to collect video, then detecting the speed of vehicle with the video has become the direction in intelligent traffic system. Because of the high accurate detecting and multi-object running after, the video speed detecting technique has get more and more attention. In this article, we detect distance with single camera. This is a sub-system of the highway video speed detecting system. The precision of the sub-system will affect the result of video speed detecting system. So what we research is how to well and truly detect displacement that the vehicle transfers on the highway with single camera?As we known, much of methods that detect distance can't apply to highway because of the large currency and complex working condition .By the experiment and the analysis of the real condition, and with the digital image technique and camera imaging model, the writer get a method that use lane to implement plane mapping then detect the distance. The basic frame of this method is that: using the lane point on the road and the lane point on the image as mapping point, then complying with the imaging model, get the relationship between road plane and image plane, and detect the displacement that vehicle transfer. Because different road have different lane, there are some roads have double lane while some roads only have single lane. For the reason, this article bring forward a method that calibrate camera firstly, then paint a virtual lane on the single lane road ,by this way ,we can detect distance either and promote the adaptability of the system. Viewing through all the frame, we can see that, the two problems in the research are detecting distance and calibration. As a result, this article emphasize on these two problems and do the following work:1. According to the condition that camera set on the highway is complex, the writer bring forward a method that use two image gray value to decide whether a pixel in the image belong to a line. This method can adapt to different climate and circumstance, the most important thing, it can detect a line accurately.2. Using the lane coordinate as character point and according to the mapping relationship between image plane and road plane, writer implements detecting moving distance of vehicle. Comparing with 3D reconstruction and other machine vision technique, the method is very simply at the same time ,it's precision is good enough.3. Because the cameras set on the highway are acute and easily to swing, so the writer use the lane as criterion to detect swinging and reduce the bad affect to this system.4. According to the situation that different road have different lane, we use a camera that have been calibrated to conform a virtual lane to resolve the situation that there is only one lane on the road.In other part of this article, we introduce the SDK and referenced device problems. In a word, the method we used in the article is effective and simply, at the same time, it has strong adaptability, and low cost. Of course, we are confronted with a lot of problem, and the construction of the system is not good enough.