Significant Obstacle Location with Ultra-Wide FOV LWIR Stereo Vision System
Yi-chao Chen,Fu-Yu Huang,Bing-Qi Liu,Shuai Zhang,Ziang Wang,Bin Zhao
DOI: https://doi.org/10.1016/j.optlaseng.2020.106076
IF: 5.666
2020-01-01
Optics and Lasers in Engineering
Abstract:Intelligent driving is an active area of research in both industry and academia. In order to overcome the shortcomings of traditional machine vision such as visibility is easily affected by illumination conditions, the blind area of infrared small field of view (FOV) is too large and could not provide depth information, this paper proposes a method for detecting significant obstacles based on ultra-wide FOV long-wave infrared (LWIR) stereo vision system. The stereo vision positioning location with ultra-wide FOV is established by the generalized fisheye camera model. On the basis of analyzing obstacle imaging scale and the structure characteristics of the proposed stereo vision system, a multi-scale salient region detection method based on composite pattern is proposed, and its implementation process is described in detail. Experiment shows that the proposed ultra-wide FOV LWIR stereo vision system is able to detect and locate significant obstacles in ultra-wide FOV and the detection rate of pedestrians and vehicles in real complex street scenes is over 92.6%. At the same time, the relative error of pedestrian positioning with a distance of 5 m to 30 m near the central FOV is between 1.6%-10.3%, and its spatial location ability and advantages are verified. The proposed stereo vision system effectively overcomes the shortcomings of existing vision systems, expands the scope of machine vision, and can be used in the field of assistant driving and intelligent driving.