Applying VGI and a Modified DBScan in Automatic Identification of Traffic Incidents

• Main Authors: WEI,CHAO-HSIEN, 魏肇賢
• Other Authors: KAO,YU-CHI
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/78183978588847071965

碩士 === 逢甲大學 === 都市計畫與空間資訊學系 === 102 === Currently in Taiwan the road traffic incidents reporting and publishing process begins with receiving phone calls from incident witnesses, followed by manual identification of incident location, confirmation of incident veracity, and then broadcasting to the public via public media, such as radio station or the internet. However, this system has many drawbacks. 1) it only covers a limited number of accidents because of inconvenient and costly phone calls. 2) incident location may be obtained inaccurately due to the diverse ways of describing a location by natural language. 3) the manual process of identification and confirmation is not quick enough for real-time traffic control or for drivers to avoid congestions. Due to the popularization of the Internet and mobile phones in last decade, the voluntary exchange and sharing of geographic information, or Volunteered Geographic Information (VGI) has exponentially increased. This research adopted the concept of VGI to collect road traffic incidents via a mobile phones application, instead of through phone calls. Upon witnessing a road incident, reporters are only required to push a button on their mobile phone with the reporting application, which automatically submits a photo of the incident and its geographic coordinates to a central database via wireless communication. With this basic incident data, an analysis of filtering and clustering is then needed in order to obtain the correct incident information in terms of location and duration, which is the major objective of this research. Several typical issues of VGI involved in the above-mentioned analysis include repeatability, timeliness and veracity. This research modified a density-based clustering algorithm, DBScan, to filter and cluster those incident data in order to solve these issues. In the modified DBScan algorithm designed by this research, the time dimension is added to the XY dimensions of the original DBScan algorithm resulting in a 3-D model (one for time and two for location) which can identify an incident and filter noises according to temporal and spatial data simultaneously. Modified DBScan runs on this 3-D model to determine the number of incidents as well as the duration of each identified incident which will be used to cancel the incident notification. Simulations of 7 incidents with purposely added noise were conducted in 3 different types of incident locations such as near intersection, in the middle of a road, and inside a traffic circle. Two performance indicators, identification rate and error rate were defined and used to evaluate the algorithm. The results suggest that the modified algorithm can successfully identify most incident with minor errors. When the searching radius parameter in the algorithm is increased, error rate also increase. The impact of noises mainly affects the determination of the duration of incidents. In general, the simulation results indicate that the 3-D model and modified DBScan algorithm developed in this research can successfully auto-identify road incidents, which may benefit the automation of road incident reporting and publishing by VGI in the near future.