中文摘要：

针对现有高速公路异常事件检测方法计算量较大，处理速度有限的问题，提出一种高速公路抛洒物事件检测与违法车辆阴影抑制算法，该算法根据视频序列中5帧图像之间运动目标位置差别很小的前提条件，间隔进行2次三帧差分，构成五帧差分；在五帧差分的基础上，在其差值之间做2次“与”运算，利用此时运动目标变化区域的灰度值设定灰度检测门限，得到前景目标图像，通过对阈值的计算辨识得到需要的参数；融合阈值自适应调整后的背景差分法，获得前景目标图像，将2帧结果图像通过“或”运算进行形态学处理后，去噪平滑得到融合后的前景目标图像，从而提取较高精度的运动目标；同时利用视觉不变性，将图像的像素值分解为色度分量和亮度分量进行阴影检测，实现对强光照下违章车辆阴影的检测和抑制。研究结果表明：该算法能兼顾处理速度和处理效果，可实现对高速公路抛洒物事件的判别和检测，并可有效处理典型高速公路异常事件，能从阴影中有效地分割出肇事车辆，处理速度可以达到25帧／s，满足实时性要求。

英文摘要：

In view of that the existing detection methods of highway abnormal events needed a large amount of calculation, and the processing speed was limited, this paper proposed a new method for incident detection of discarded things in highway and the shadow suppression algorithm for illegal vehicles. The algorithm used the preconditions of small difference in moving target position among 5-frame images of video sequence, and 2 times 3-frame difference were conducted at intervals to form 5-frame difference. On the basis of this, foreground target image could be got through twice ＂and＂ operation with the difference and gray scale detection threshold. The required parameters were obtained by calculation and identification of the threshold which was set by using gray value of the moving target changing area. Background-difference method of fusion threshold self-adaptation was used to obtain foreground target image. The two resulting images was processed using morphologic operator by ＂or＂ operation, and the fused foreground target image was obtained by denoising smooth so as to extract the moving target with high precision. Taking advantage of the visual invariance, the pixel values of the image were decomposed into luminance comp detection and suppression of the s show that the new algorithm can accident vehicle can be effectively can reach 25 frames per second to onent and chrominance component for shadow detection, and hadow of vehicles under strong light was realized. The results effectively deal with typical highway abnormal events, and the segmented from the shadow. Moreover, the processing speed meet real-time requirements. 5 figs, 25 refs.