露天场景下高速公路违章停车视频检测算法

Detection Algorithm for Highway Illegal Parking in Open-air Scenes

针对现有的停车检测算法存在不能很好地适应露天场景环境变化以及误检率过高的问题,设计了一种新的违章停车视频检测算法;利用自适应混合高斯模型实现场景背景提取与更新,获得可能的运动前景目标,通过基于像素时间序列特征的稳态分析法来检测静止目标,并根据对象区域特征实现停驶车辆的辨识,从而准确的检测出违章停车;实验结果表明,提出的算法在高速公路露天场景中正检率达到90%以上,与其他停车检测算法相比较,在减少计算量、提高系统实时性的同时,具有较好的正检率。

To overcome the problems of existing parking detection algorithms, not adapt to the environmental changs of open--air scene and high false positive rate, a new video based algorithm is developed to detect the illegal parking. An adaptive mixed Gaussian model is used to extract the background in the complicated environment. The still objects are detected by the time series {eature of the pixel level. The still vehicles are recognized by the regional feature of the object level. And thus the illegal parking is accurately detected. The experimental re- sults that the proposed method can reach more than 90% of the positive detection rate. Compared with the other algorithms, it not only reduces computations and improves the system real--time, but also with better accurate detection rate.