视频监控场景下行人衣着颜色识别

Clothing color recognition of pedestrians in video surveillance

近年来,由于视频监控在各地安防的广泛应用,行人的精细化识别显得尤为重要,其中行人的衣着颜色是最显著的外观特征,其识别的正确性直接影响视频检索中对特定行人的检索.论文提出了一个简单实用的行人衣着识别系统,可以有效地识别行人衣着颜色.首先,结合HOG(histogram of oriented gradient)算法和Grabcut算法自动地对监控图像中的行人进行精确分割;然后,在利用外观划分模型精确地分割出行人的上身和下身后对上下身分别分割成若干个小块;最后,使用KNN(k-nearest neighbor)分类方法判断每个块的颜色,通过所有块的颜色标签投票决定衣着颜色.最终,使用收集的监控视频图像数据集验证此方法的有效性和实用性.

In recent years,video surveillance is widely used in country security,making finegrained pedestrian recognition becomes important.In particular,the clothing color is the most salient feature,and its recognition accuracy directly influences the retrieval accuracy of specific pedestrians in video retrieval.This paper presented a simple but fast system of pedestrian clothing recognition,which could effectively identify pedestrian clothing color.Firstly,through combining HOG and Grabcut algorithm,the pedestrians in frames of surveillance could be accurately segmented.Then,we put forward the appearance of partition model which was simple and effective segmentation pedestrian tops and bottoms,and used KNN classification method to determined the color of each patch,through all the patch color vote to decided what kind of clothing colors.Finally,experiments were carried out in this collection of surveillance video images data sets,which had verified the validity and practicability of this method.