基于相机和红外反射率的激光雷达感知的增强

ENHANCEMENT OF LIDAR SENSING BASED ON CAMERA AND INFRARED REFLECTIVITY

周钰

11930217

硕士

材料工程

洪小平

2021-05-24

2021-06-15

南方科技大学

深圳

测绘行业自古以来就对国计民生意义重大。传统的测绘手段耗费大量人力物力，且环境艰苦， 存在安全隐患。但是近年来人们对真实物体三维的电子模型需求越来越旺盛，例如说基于虚拟现实的影音娱乐、基于增强现实的旅游等都需要对场景的真实重现。现今的三维重建的方案一般使用激光雷达，但是它只能获取空间位置信息， 本文的创新点在于提出了一种低成本的方案，创造性得将工作在可见光波段的相机和工作在红外波段的激光对不同材料的反射率这两者结合起来， 来增强激光雷达的感知能力，在此基础上建立真实环境的三维电子模型或者地图。相机作为视觉传感器可用来增强激光雷达的感知能力。相机能够获取400-780nm 波段内的可见光信息，反映出物体的颜色，颜色信息能够在一定程度上反映物体的材质， 通过图像处理能够对不同物体进行分割、识别和分类，因此可以用来增强激光雷达的感知能力。激光对不同材料物体的反射率信息也可以用来增强其感知。激光雷达发射红外波段的激光，9 05n m 的红外激光对人眼的安全阈值较高， 且成本低廉，相对稳定， 因此常常用于激光雷达。反射率信息反应了入射激光和反射激光的功率之比，一般而言， 反射率与激光的波长、温度等有关， 还有物体表面材质有关。光滑明亮表面的反射率就比粗糙黑色表面的物体更高。镜面反射物体比如金属、陶瓷等反射率就比漫反射物体例如墙壁、白纸等要高，因此反射率可以作为识别物体特性的另一维度。在点云配准时可以作为除x、y、z 坐标之外的第四维度信息提高配准的准确性。反射率在材料识别和激光点云配准方面增强了激光雷达的感知能力。根据上述两点， 本文针对只带有红外反射信息的激光雷达点云和带颜色信息 的相机图片在现有算法基础上进一步展开了研究，利用激光反射率信息 对不同姿态下的激光点云进行配准和拼接，激光雷达和相机的标定结果对 点云进行上色。最终可以得到真实场景的电子重建模型，基于此，可以实 现很多有趣且意义深远的应用，比如说智能校园的导航、校园建筑的规划、便利校园的建设等等。

The surveying and mapping industry has been of great significa nc e tothe national economy and the people's livelihoo d since ancient time s .Traditiona l surveying and mapping methods consume a lot of manpow e rand material resources, and the environmen t is di fficult and there arehidden many safety risks. H ow ever, in recent years, people's demand forthree-dim en si ona l electronic models of real objects has become more andmore vigorous. For example, audio -visua l entertain m en t based on VR(virtual reality ), tourism based on A R (augmented reality ) require the rea lreconstruc t ion of the scene. Today's 3D reconstruc ti on schemes genera l l yuse lidar (light detection and ranging) , but it can only obtain spat i a lposition informat ion . The innovation of this article is to propose a low -cost scheme that creatively combines a camera that w orks in the visi b l elight band and a lidar that w orks in the infrared band. The reflecti vi ty ofdifferent materia ls is combined to enhance the sensing ability of lidar, andon this basis, a three-dim en s ion a l electronic model or map of the rea lenvironmen t can be established .The camera can be used as a visual sensor to enhance the percept i o nability of lidar. The camera can obtain the visible light informati on in the400-780nm band and refle ct the color of the object. The color informa t i o ncan reflect the material of the object to a certain extent. It can segme n t ,identify and classify different objects through image processing, so it canbe used to enhance the sensing ability of li dar.The reflecti v it y informat ion of the laser on objects of differ e n tmaterial s can also be used to enhance its perception. Lidar emits lasers inthe infrared band. The 905nm infrared laser has a high safety threshold forthe human eye, is low in cost and relativel y stable, so it is often used inLidar. The reflectanc e informatio n reflects the ratio of the pow er of theincident laser to the reflected laser. G enerally speaking, the reflectanc e isrelated to the w avelength and temperatur e of the laser, as w ell as the surface material of the object. The reflectiv i ty of a smooth and brig h tsurface is higher than that of an object w ith a rough black surfa c e .Specular reflectio n objects such as metals and ceramics have high e rreflectiv i ty than diffuse reflectio n objects su ch as w alls, w hite paper, etc .Therefore, the reflectiv i ty can be used as another dimension to ident i f ythe character is t ic s of the object. When the point cloud is registered , it canbe used as the fourth -dimen s ion a l informat ion in addition to the x, y, andz coordinates to improve the accuracy of the registra ti on. Reflect a n c eenhances the perception ability of lidar in material recognitio n and lid a rpoint cloud registra t ion .Based on th e above tw o aspects, this thesis further studies the lida rpoint cloud w ith only infrared reflecti on informat ion and the camera ima g ew ith color informati on based on the existing algorith m, and uses the las e rreflectiv i ty informat io n to register and map lidar point clouds in diffe r e n tposes and color point clouds by the calibrati on results of lidar and camer a .In the end, an electroni c reconstruc t io n model of the real scene can beobtained. Based on this, many interest ing and far -reach ing applicat i o n scan be realized, such as the navigati on of smart campuses, the planning ofcampus buildings, and the convenience of campus construct io n.

测绘 激光雷达 相机 红外反射率 点云配准

surveying and mapping lidar camera infrared reflectivity point cloud registration

中文

独立培养

南方科技大学

周钰. 基于相机和红外反射率的激光雷达感知的增强[D]. 深圳. 南方科技大学,2021.