Analysis of a Distance Measurement System Based on Monocular Camera
Rodrigo Adamshuk Silva1; João Henrique Zander Neme1; Erick Margraf1; William Takeshi Omoto1; Felipe Rezende Franco1; David Carvalho Andrade1; Felipe Bueno1; Rodrigo Amaral2; Artur Costa Ventura2; Sérgio Okida3
1 UTFPR - Campus Ponta Grossa; 2 FCA; 3 UTFPR-Campus Ponta Grossa
doi:10.20906/CPS/CILAMCE2017-1100
Resumo
In 2016, according to the IPEA report "TRAFFIC ACCIDENTS ON BRAZILIAN FEDERAL HIGHWAYS", the rear-end collision and the trampling of people represented, respectively, 29.2% and 14.6% of the accidents and has been responsible for the almost 8.4% and 2.5% of injuries and deaths. Collision mitigation is an automotive function that enables to measure the distance of an object and alerts the driver to take preventive measures and avoid collisions, thereby reducing the number of deaths, injuries and material damages. This work present a strategy able to evaluate an ahead vehicle distance based on monocular camera installed in a vehicle windshield. The purposed system consists of the stages of image acquisition, preprocessing, selection of the region of interest, object detection and distance determination. Active contour model is the method employed to detect the vehicle. To measure the distance with a monocular camera is necessary to correct the perspective effect produced by image acquisition process, a distortion in which parallel lane markings converge to a same point on the horizon. The Inverse Perspective Method (IPM) corrects this distortion in a way that each pixel represents road portions with same dimensions. Afterwards calibration, the distance in pixels between the detected vehicle and the camera is converted to meters. The distance measurement system using IPM was evaluated for a flat highway, with a vehicle positioned at distances between 5 and 50 meters from the camera, the measurements were taken every 5 m. The images have been acquired in 3 periods of the day, in the early afternoon, in the late afternoon and in the evening. For each vehicle positioning, 500 measurements were taken. The mean error obtained ranged from 0.04% to 16.36% of the value accepted as true. By analysis of variance method (ANOVA) with p = 0.05 it was verified that the average of the distances calculated in the 3 periods of the day are significantly different (p<<0.05), that is, the environmental conditions of luminosity influence
Palavras-chave: monocular camera; distance measurement system; Inverse Perspective Method