基于视频检测技术的隧道停车检测与识别算法

A Detection and Recognition Algorithm of Tunnel Parking Based on Video Processing

针对现有视频停车检测算法多采用背景逐渐更新的方式,不能较好适应于存在照明突变和渐变的高速公路隧道交通场景的问题,设计了一种能有效抑制隧道光照变化的隧道停车检测与识别算法;首先针对光线变化等因素对隧道背景的影响,采用非参数核密度模型进行背景抽取,并结合相应的背景更新算法与背景差法实现运动前景的提取;然后改进基于灰度分析的时间序列法,通过叠加及归一化的方法充分利用前景图像二值信息进行像素级时间序列分析,以获取静止前景目标,并通过连通域分析滤除其他小面积静止物体的干扰;最后运用多轮廓搜索算法,针对检测区域内不同分区中的静止车辆,进行对象级多特征辨识,并依据一定的报警机制标识静止车辆,实现隧道停车事件自动报警;实验结果表明,提出的算法能有效抑制隧道光照变化,较好地实现了隧道停车的自动检测与识别,具有较高的准确性和有效性。

To overcome the problem that the existing video parking detection algorithms which mostly update background gradually can not better to adapt to the highway tunnel traffic scenes where exist illumination abrupt changes and gradations, a detection and recognition al- gorithm of tunnel parking based on video processing which can effectively inhibit tunnel illumination changes is proposed. Firstly, aiming at the influence of illumination changes and other factors to the tunnel background, nonparametric kernel density estimation is used to build the background model. Then, the moving foreground is extracted by combining with the corresponding background update algorithm and the background subtraction. Next, in order to obtain the static object, the binary information of foreground images is made full use of to perform a pixel level temporal sequence analysis by the stack and normalization method which improves the temporal sequence method based on gray value analysis. In addition, the interferences of other small area static targets are filtered by the connected field analysis. Besides, the still ve hicles in different subareas of the detection area are recognized by the applications of multi--contours search algorithm and the regional multi --feature of the object level. Finally, according to a certain alarm mechanism, the marking of still vehicles and the auto--alarm of tunnel parking event are achieved. The results of experiment validate the effectiveness and accuracy of the algorithm.