| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊管理學系 | zh_TW |
| DC.creator | 高明郁 | zh_TW |
| DC.creator | Ming-Yu Kao | en_US |
| dc.date.accessioned | 2020-7-29T07:39:07Z | |
| dc.date.available | 2020-7-29T07:39:07Z | |
| dc.date.issued | 2020 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=107423012 | |
| dc.contributor.department | 資訊管理學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 隨著物聯網和5G的技術日漸成熟，自駕車相關的研究和應用也隨之迅速地發展，目的為讓自駕車可以在變化快速的交通環境中安全行駛，降低交通事故的發生率，但由於目前自駕車受到各方面發展的限制，因此，仍多以駕駛者根據系統之預警自行做出決策。
為了使自駕車能在短暫的時間做出更正確的判斷，以降低發生碰撞的機率，系統以自駕車為中心的360度全景偵測範圍，依行駛速度動態地調整範圍大小，偵測範圍可劃分為「警示區域」和「鄰近區域」，利用LSTM-CNN的混合式架構分別對兩個區域的車輛及物體之間的動態交互作用進行軌跡預測，再將自駕車警示區域的車輛及物體預測後的軌跡，進行所需碰撞時間的計算，並且根據自駕車的行駛速度調整所需預警的碰撞時間 (Time to Collision, TTC)，以減少碰撞預警時間過早或不及的狀況。
若需確立防撞系統的安全性，則需使模型預測軌跡具有相當的準確性，因此，本研究採用擁有複雜且異構交通環境的Argoverse Motion Forecasting v1.1的部分資料集，並以TraPHic Chandra, Bhattacharya, Bera, & Manocha (2019)為參考方法進行比較，實驗證明在評估指標最終位移誤差(Final Displacement Error, FDE)和平均位移誤差(Average Displacement Error, ADE)皆較TraPHic佳，代表提出的方法能夠更加準確地預測軌跡，因此，將此預測值作為計算所需碰撞時間的輸入，能夠即早提供具有可靠性的碰撞預警。 | zh_TW |
| dc.description.abstract | As Internet of things and 5G technology are gradually mature, researchs related to automatic vehicle are also growing rapidly. The purposes of those research and applications are to make automatic vehicle drive safely in fast-changing traffic environment and reduce traffic accidents. However, due to various aspects of limitations, drivers still make their own decisions based on system warnings in automatic vehicles.
In order to let Self-driving make correct decisions in short time to reduce the probability of collisions, the system takes the self-driving as the center of the 360-degree panoramic detection range. Besides, based on driving speed to dynamically adjust range size, warning area and nearby area are divides. Using the hybrid architecture of LSTM-CNN to predict the trajectory of the dynamic interaction between vehicles and objects in the divides areas, and then calculating the required collision time by the vehicles and the predicted trajectories of objects in self-driving warning area. Still, according to the speed of self-driving, Time to Collision for warning can be adjusted so as to reduce the situation that the collision warning time is too early or too late.
If the safety of collision avoidance system needs to be established, the model with accuracy is necessary to be made to predict the trajectory. Thus, the dataset of Argoverse Motion Forecasting v1.1 is adopted with complex and heterogeneous traffic environment, and Traphic is used as reference to execute comparison. According to the experiment, Final Displacement Error (FDE) and Average Displacement Error (ADE) of evaluation criteria are better than the method of TraPhic, means the method of this study can predict the trajectory with higher accuracy. Therefore, using the predicted value as the input for calculating the required collision time can provide more reliable and early warning of collision. | en_US |
| DC.subject | 自駕車 | zh_TW |
| DC.subject | 避免碰撞 | zh_TW |
| DC.subject | 軌跡預測 | zh_TW |
| DC.subject | 動態交互 | zh_TW |
| DC.subject | Self driving | en_US |
| DC.subject | Collision Avoidance | en_US |
| DC.subject | Trajectory Prediction | en_US |
| DC.subject | Dynamic Interaction | en_US |
| DC.title | 自駕車全景動態偵測防撞系統 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Collision Avoidance System of Automatic Vehicle Panorama Dynamic Detection | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |