高速公路上鄰近車輛之危險動向偵測
No Thumbnail Available
Date
2003
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
本篇主要為應用影像技術偵測在高速公路上行駛時鄰近我車之車輛的危險動向。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，主要針對為鄰近我車之車輛；知覺分析器則是利用—STA (spatial-temporal attention)類神經網路模組來記錄鄰近車輛之移動方向，其結果稱為注意力圖像（attention maps），隨後我們將此圖分割為五個視窗，以便偵測不同位置的障礙物，對於每個視窗我們計算其偏態（skewness）特徵值，作為分類時的輸入值；概念分析器則是根據各個視窗計算出來的偏態值利用CART （configurable adaptive resonance theory）類神經網路來做分類。最後在決策（decision making）模組中應用模糊理論整合各個CART類神經網路的結果以輸出最後分類的結果。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 behaviors of the nearby vehicles on the expressway. The system consists of three components: sensory analyzer, perceptual analyzer and conceptual analyzer. The sensory analyzer detects moving objects, especially the nearby vehicles. The perceptual analyzer records the motion direction of nearby vehicles in the attention map of a STA(spatial-temporal attention) neural network. We divide the attention maps into five overlapping windows from each of which the feature of skewness is computed. Each feature is fed into a CART (configurable adaptive resonance theory) for motion classification. Using the fuzzy integral, individual decisions made by separate CART neural networks are combined and the final decision is attained. A number of experimental results are presented, which revealed the feasibility of the proposed approach.
We propose a system which can detect the motion behaviors of the nearby vehicles on the expressway. The system consists of three components: sensory analyzer, perceptual analyzer and conceptual analyzer. The sensory analyzer detects moving objects, especially the nearby vehicles. The perceptual analyzer records the motion direction of nearby vehicles in the attention map of a STA(spatial-temporal attention) neural network. We divide the attention maps into five overlapping windows from each of which the feature of skewness is computed. Each feature is fed into a CART (configurable adaptive resonance theory) for motion classification. Using the fuzzy integral, individual decisions made by separate CART neural networks are combined and the final decision is attained. A number of experimental results are presented, which revealed the feasibility of the proposed approach.
Description
Keywords
動向偵測, 車輛偵測, 模糊整合, Motion Detection, Vehicle Detection, Fuzzy Integral