高速公路上鄰近移動車輛之動向偵測

Motion Detection for Nearby Vehicles

本篇主要為應用影像技術在偵測在高速公路上鄰近車輛的動向。系統主要分為三個部份：感覺分析器（sensory analyzer）、知覺分析器（perceptual analyzer）與概念分析器(conceptual analyzer）。感覺分析器可找出影像中移動的物體，主要為鄰近的車輛；知覺分析器則是利用STA（spatial-temporal attention）類神經網路模組繪出移動的鄰近車輛之移動方向，也就是注意力圖像(attention maps），並將所得到的注意力圖像分割為五個視窗，以便偵測到不同位置的障礙物；概念分析器則是針對各個視窗分別利用CART（configurable adaptive resonance theory） 類神經網路來做分類。最後利用決策產生器模組中的模糊理論整合各個CART類神經網路的分類以輸出分類結果。在實驗結果中，我們提出數個例子以驗證我們的方法。

We propose a system which can detect the motion patterns of nearby vehicles on an expressway. The system consists of three components: sensory, perceptual, and conceptual. The sensory component can extract the spatial and temporal information about the moving objects, including nearby vehicles. The perceptual component uses a STA（spatial-temporal attention） neural network to form "attention maps" which indicate the motion direction of nearby vehicles. We divide these attention maps into five overlapping windows in order to analysis more then one vehicle motions. We extract skewness from each window. These features are fed into their corresponding CRAT（configurable adaptive resonance theory） neural network in conceptual component for classification. Using a modify fuzzy integral technique, individual decisions are made by each CART neural network, and then these decisions are collected to make the final decision about the motion class of the moving vehicle.