即時車輛和行人偵測與顏色辨識系統

Abstract

在視訊安全監控的相關研究中車輛與行人偵測和顏色辨識是相當重要的議題。 在 本論文中，我們將針對車輛與行人的偵測和顏色辯識技術進行探討。 在車輛行人 偵測方面，目前的方法大多使用2D資訊做為特徵，例如邊緣、顏色、輪廓、動 作...等。 其中只有少部分的研究使用3D的特徵。 本論文提出一套新的演算法使 用3D資訊來偵測車輛和行人。 首先會使用背景模型的技術來取得前景移動物體， 對於每一個前景移動物體，我們會利用相機的內外在參數來計算物體在3D空間中 的大小。 我們使用calibration-free的方法來估測攝影機參數， 其方法只需要在場 景點選長方體的六個點即可。 顏色辨識系統方面，我們會利用Bayesian分類器來 訓練所定義的顏色在HSV色彩空間的決策邊界, 然後依據車輛和行人影像中的像素 在所定義的顏色區域的分布來決定其顏色。 經由實驗結果，所提出的方法都能有 效的運作

We propose a real-time intelligence surveillance system.Two important topics are studied, including vehicle and pedestrian detection, vehicle and pedestrian color classification. Existing pedestrian and vehicle detection algorithms utilize 2D cues of objects, such as pixel values, color and texture, shape information or motion. Some of them require heavy computation power and are thus prohibited from real-time applications. While many researchers focus on modeling objects based on 2D cues, the use of 3D cues in object detection are not well studied. In this paper we propose an algorithm that utilizes 3D cues to perform pedestrian and vehicle detection. The 3D cues of objects in a static scene monitored by a camera can be obtained using the intrinsic and extrinsic parameters of that camera. We apply a calibration-free method to estimate the camera parameters. This method simply requires users to specify 6 vertices on a cuboid in the scene. In the aspect of vehicle color classification, we use Bayesian classifier to trained the decision boundaries of defined color in the HSV space, then determining the color of the object according to distribution of the the pixels in the vehicle and pedestrian images on the defined color region. Experiment results demonstrate our proposed method can work efficiently.