基於Wi-Fi通道狀態資訊之車輛盲點偵測系統

Abstract

隨著道路環境日益複雜、擁擠，盲點偵測已成為現今車輛普遍會配備有的駕駛輔助系統。然而，它們在實作上大多仍採用昂貴的感測器，如攝影機、光達、雷達等。基於車聯網系統模型，本研究提出結合無線網路感測技術，以 Wi-Fi 通訊時所產生的通道狀態資訊 (Channel State Information) 來實現盲點偵測系統的功能，以期達到降低成本、資源有效再利用的目標。 為驗證此方法的可行性，我們將 Wi-Fi 通訊設備及光達佈署於車輛，利用它們於實際行駛過程中所收集的感測資料來建構資料集、實行特徵工程、訓練深度學習模型。模型設計以固定時間間隔進行二元分類，判斷當下訊號源是否位於盲點區域內，並對照實際情況來計算預測的正確率。最終的結果顯示，此系統可以在有限的頻寬及運算資源條件下達到超過 0.9 的 F1-score，足以應付正常使用情境。

As traffic environments get increasingly complex and congested, Blind Spot Warning (BSW) system has become a common feature in modern vehicles. However, its implementation typically relies on expensive sensors such as cameras, LiDAR, and radar. Based on Vehicle-to-Vehicle (V2V) communication system model, this study proposes a more cost-effective approach that incorporates Wi-Fi sensing techniques to realize its safety functionalities. The main objective is to efficiently reuse existing communication resources by leveraging Channel State Information (CSI) from signals. To verify the feasibility of this design, we conducted real-world experiments with Wi-Fi communication devices and LiDAR deployed on vehicles. The collected measurements were then utilized to construct CSI datasets, conduct feature engineering, and train a deep neural network (DNN). We designed the model to perform binary classification at fixed time intervals, predicting whether the signal source is currently located within the blind spot area. Ultimately, the evaluation results showed that the proposed system can achieve an F1-score of over 0.9 under limited bandwidth and computational power, which is sufficient in general use cases.