Vehicle speed analysis from video considering scanning time

Abstract

Recently, the importance of vehicle speed analysis from video images in traffic accidents and incidents has been increasing, and with the development of camera technology, there is an urgent need to improve image analysis technology. Among the influencing factors, the complexity of the frame rate due to digitization of video signals, and the time difference of imaging within the same frame due to the rolling shutter method of CMOS cameras are thought to influence the accuracy of vehicle speed analysis.

Therefore, we developed a software that can automatically read frame rates down to 1 ms by analyzing video images of a device (Time Counter) with a stopwatch function using LEDs. This software automatically read frame rates by cropping the frames, correcting lens distortion, generating orthogonal images through homographic transformation, and binarization of the images, then reading the coordinates of the lit LED points. Next, we developed a scanning time measurement device. This is a device that blinks an LED tape light at an arbitrary frequency. When this device is shot by a CMOS camera with rolling shutter method, the LED tape light is imaged in stripe pattern. The scanning speed can be measured by measuring the width of the stripe.

Moreover, we proposed a vehicle speed analysis method that takes scanning time into account, and conducted physical experiments to confirm its effectiveness．As a result, the proposed method achieved an error of ± 0.1 km/h, while the conventional method showed an error ranging from ‒1.48 km/h to +1.46 km/h. The proposed method improved the accuracy of vehicle speed analysis.