基于知识和外观方法相结合的后方车辆检测

单独应用基于知识的方法或者单独应用基于外观方法检测是否存在车辆有一定的局限性,因此提出将二者结合起来用于静态图像后方车辆的检测.首先,利用分割算法获得感兴趣的区域(region of interest,ROI),利用基于知识(如车底阴影、颜色等信息)的方法,将被确认为是非车辆(背景)的ROI过滤掉,然后再对过滤后的结果应用基于外观的方法进行车辆检测.在不同的道路(高速公路、城市普通道路和城市窄道)条件以及白天不同光照条件下对车辆进行检测,结果表明,该算法的识别可靠性更高,适应性更好.

Rear-Vehicle Detection Combining Both Knowledge-Based and Appearance-Based Methods

There is unavoidably a limit to either knowledge-based or appearance-based methods when using any of them singly to detect the existence of vehicles.An algorithm combining both of them is therefore proposed to detect rear-vehicles in static images.First,the ROIs(regions of interest) obtained from segmentation algorithm are filtered which are regarded as belonging to background by using knowledge-based methods such as the shadow underneath a vehicle and color information.Then,the vehicles are detected with appearance-based methods on the remains.The detection results of vehicles traveling on highways,urban common roads and urban narrow roads under various illumination conditions on daytime indicated that the proposed algorithm has better reliability and higher adaptability than either of the algorithms singly based on knowledge or appearance.