本文提出一種基於視覺分析的車道變換輔助系統。此系統以行車紀錄器所得影像偵測後方車輛，判定接近車輛的威脅程度及是否可安全變換車道。為了獲得好的車輛偵測效能，系統中使用深度學習神經網絡。不過，在不佳的天氣（如雨和霧）下會導致系統性能的下降。實驗結果顯示，運用除去雨和霧的影像復原技術可提升系統性能。

A lane change assisting system based on visual analysis has been proposed to improve driving safety in this paper. This system detects the rear vehicle with the images obtained from a dashboard camera, determines the threat level from the approaching vehicle when changing lane, and makes a lane change suggestion to drivers. To have better vehicle detection results, we use a deep learning neural network in the system. However, the images obtained in a bad weather, say rainy and foggy, make vehicle detection difficult for our system, which may result in poor system performance. Some image restoration techniques for removing rain and fog effects are adopted in this study to deal with this issue. The experimental results show that applying the techniques indeed improves the system performance in rainy and foggy conditions.