Initiative Anti-collision Monitoring System For Loaders Based On Stereo Vision

Taking into account the prone collision of construction machinery when collaborating in groups, it was of great importance that active anti-collision technology of loaders had achieved. Combining the National Science and Technology Support Program named by " Engineering Machinery And Equipment Research and Development For An Integration of Intelligent Energy-saving And Safety " and using XG958H loaders for the application object, the paper presented a new stereoscopic3D reconstruction method that used a binocular vision sensor and laser sensors for3D reconstruction of the actual operating scenarios of loaders. According to an analysis of a3D scene graph integrated by a natural scene graph with a depth graph to achieve the purpose of monitoring alarm. Finally the method was applied to active anti-collision monitoring system of loaders.The main work of this paper include the following.First on-site calibration method of the active anti-collision system was studied. The calibration method was divided into three steps:Firstly, the left and right cameras are calibrated to get its internal parameters.Then, a conversion formula was used to calculate the relative pose between two cameras. Finally, transformation relationships between a laser scanning point coordinate and the image coordinate was built.Then3D information acquisition of the natural scene method was studied.3D information acquisition method was divided into three steps.First, characteristics region and non-characteristics region were determined by feature point detection and zoning. Secondly, local depth graph of feature region based on binocular vision was obtained through feature point detection and matching,3D coordinates calculation and correction and image registration. Finally, the depth graph based on laser scanning was obtained through laser scanning platform scanning non-feature regions and fitting3D information of laser spot. Finally, a3D scene graph fusion method was proposed.3D scene graph fusion method was divided into three steps.Firstly, all the targets areas were obtained by "region growing" segmentation method using for natural scenes graph. Then, depth image of object within feature region was generated through finding the right matching gray value and using the depth image based on binocular vision to fill up regions of it. The depth image of object within non-feature region was generated through using depth image based on laser scanning to project regions of it. Finally the3D scene graph relative to nature scene graph was generated by an use of "plus operator " between depth images of all targets.There are two aspects for innovation. On the one hand, the3D scene graph relative to nature scene graph was obtained by filling up characteristic regions in diffusion and filling up non-feature region in projection. On the other hand, binocular vision sensors and laser sensors were fused to obtain a3D scene graph and active anti-collision purposes was achieved by an analysis of the3D scene graph.