我們每個人每天或多或少會製造出一些垃圾，而在台灣，垃圾通常透過垃圾車運送至垃圾焚化廠進行處理。垃圾在焚化廠內處理的第一步為將垃圾貯坑內的垃圾，經由天車控制人員操控吊車，夾取貯坑內的垃圾至垃圾進料斗來進行後續的垃圾處理。本論文為減輕天車操控人員的負擔，開發出一套垃圾貯坑的處理系統，使用二維光達掃描貯坑內垃圾並輸出三維點雲，再根據垃圾的三維點雲決定垃圾夾取的位置與順序，夾取後再重新掃描夾取周圍的垃圾深度，建立新垃圾的三維點雲，再重複夾取過程，一直循環下去，直到垃圾量夠少才停止。以上用以大量取代操控人員的工作，使得垃圾處理更即時、更自動、更有效率。另外，還可透過垃圾的三維點雲建模，將垃圾山模型重新置入模擬環境當中，作為天車操控人員垃圾輪廓的參考。 本論文的研究分為四大部分，第一部分為使用Blender軟體製作垃圾模型放入Gazebo模擬環境，並在Gazebo模擬環境進行天車的模擬，將模擬之二維光達安裝在模擬環境中天車的側面，透過程式控制天車的移動，來對整個垃圾貯坑進行垃圾深度的掃描，並在最後進行整個垃圾貯坑的深度圖呈現。第二部分會針對掃描完整個貯坑後的點雲進行後處理，過濾掉不需要的深度資料點，對剩餘的深度資料點進行接續的夾取點判斷，並規畫出一套夾取順序的流程。當垃圾經由多次被夾取後，進行被夾取範圍內深度資料的更新，最終目的為將原先放置於Gazebo模擬環境內貯坑的垃圾，盡可能的去除。第三部份針對點雲建模的部分，將光達掃描的原始掃描資料點，進行補中值點的動作以增加點雲稠密性，接著將填補完中間值的點雲，透過CloudCompare軟體來建模出一個三維立體模型。第四部分為設計一個真實的小型垃圾貯坑場域，安裝二維光達在由馬達控制的移動平台上，透過實際的模擬，確認演算法是否得以正常的運行並呈現出正確的深度資料投影點。如此的模擬，為希望當二維光達被安裝於真正的垃圾天車上時，設計出來的演算法能夠發揮其功能與效率。 ;Each of us generates some amount of waste every day. In Taiwan, garbage is usually transported by garbage trucks to incineration plants for processing. The first step in the incineration process involves using a crane controlled by operators to grab the garbage from the refuse bunker and transfer it to the refuse charging hopper for further processing. In this thesis, a refuse bunker processing system is developed to alleviate the burden on the crane operators. It utilizes a 2D lidar to scan the garbage in the refuse bunker and export a 3D point cloud. Based on the 3D point cloud, the system determines the position and order of the garbage-grabbing points. After grabbing, the system rescans the depth of the surrounding grabbing points to create a renewed 3D point cloud. This process is repeated until the garbage is reduced to a low level. The system aims to replace manual labor with automation, making garbage processing more real-time and efficient. By modeling the garbage from the 3D point cloud, the model will be put into the simulation environment, serving as a reference for crane operators in determining the contour of the overall garbage. The research in this thesis consists of four main parts. The first part involves creating garbage models using Blender and placing them into the Gazebo simulation environment. The simulation of the crane is conducted, and a 2D lidar is installed on the side of the crane. The crane′s movement is controlled by the program to scan the depth of the entire refuse bunker and a depth map of the entire refuse bunker is generated. In the second part, post-processing is performed on the scanned point cloud of the entire refuse bunker. Unnecessary depth data points are filtered, and the remaining depth data points are used to determine the grabbing points. As the garbage is grabbed multiple times, the depth data within the grabbing range will then be updated. The ultimate goal is to remove as much garbage as possible from the refuse bunker. The third part focuses on point cloud reconstruction. The original scanned point cloud is processed to fill in missing values and increase the density of the point cloud. The filled-in point cloud is then used to create a 3D model using CloudCompare. The fourth part involves designing a real small-scale refuse bunker. A 2D lidar is mounted on a mobile platform controlled by the motor. The algorithm′s functionality is verified through actual simulations in projecting correct depth points. These simulations aim to ensure that the designed algorithm can perform effectively and efficiently when the 2D lidar is installed on an actual garbage crane.
