Dynamic Searching Approach for the Fine Particulate Matter (PM 2.5) Pollutant Source using Drone

• Main Authors: Lin, Yen-Cheng, 林彥程
• Other Authors: Ho, Yao-Hua
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/98nz8n

碩士 === 國立臺灣師範大學 === 資訊工程學系 === 107 === Living in an environment filled with air pollution will affect our body health, cause chronic illness, and increase the fatality rate. However, it is difficult to regulated air pollution effectively, due to the need to accurately gathering evidence to prove any illegal emission. In this research, we propose a method to exploit the ability of a drone to locate air pollution (i.e., particulate matter) emission source quickly. Utilizing the information provided by an existing sensor network, the drone is able to make correct decisions when searching for pollution sources. In the proposed system, Location Aware Sensing System (LASS) provides the continuous monitoring information of PM 2.5 (Particulate Metter 2.5) to initialization searching plan by limiting a searching area. In the beginning, our drone utilizes the PM2.5 concentration information provided by LASS to adjust its searching direction and distance to an intermediate point. After an intermediate location point is reached, our drone will stop and sense the current PM2.5 concentration. Next, the drone continues to adjust the searching path with its searching direction and distance when the concentration level increased, respectively. The three searching path strategies are proposed - Greedy, Dynamic, and Hybrid Approach. The searching process repeats itself until ten of the continuous sensed PM2.5 concentration levels dropped below a threshold or the power of drone fell lower than the maximum flight time (with the reserved power for the return home distance). Once the searching process is finished, the location of the air pollution emission source is estimated by the highest contraction level measured by drone. The three of proposed strategies are compared with Random Way-Point and Space-Filling Curve. Experiment results show our proposed techniques are able to achieve estimation error below 2 meters within 20 minutes.