Abstract: Video-based traffic flow detection systems are easily influenced by background changing and vehicle shadows. A method for traffic flow detection using self-adaptive background difference and shadow removing is proposed. First, the adaptive background model is constructed and used to extract image background; and then interested vehicles are detected from video candidate area by self-adaptive background difference. Changes the difference image into binary image by given thresholding, and fill the holes within the extracted objects using morphological reconstruction by erosion. Second, according the fact that the gray value of the shadow area is less than that of the vehicle area, sweep the binary image object along the direction from shadow area to vehicles area, and compare the conjoint pixel gray value of the original gray image in the same position. By this way, most vehicle shadows can be removed. Simulations show that this method can efficiently detect the highway traffic flow and is less influenced by vehicle shadows and background changing.

Key words: vehicle flow detection, candidate area, self-adaptive background difference, shadow removing

摘要： 针对视频车流量检测容易受背景以及车辆阴影等因素影响的问题，提出了一种自适应背景差分结合阴影去除的车流量检测方法。首先，建立自适应背景提取模型；然后，利用差分法从视频检测区域提取包含阴影的车辆目标，并进行二值化处理和孔洞填充；接着依据阴影区域相对于车辆区域灰度较小的特点，从填充后的二值图像阴影区域向车辆区域方向进行像素值比较，从而检测并去除阴影；最后，通过设定两排检测窗口进行车流量计数。实验结果表明，该方法受背景和车辆阴影等影响较小，在不同气候环境下具有较高的车流量检测准确率。

关键词: 车流量检测, 检测区域, 自适应背景差分, 阴影去除