基于热点区域定义的人数统计方法研究

发布时间：2018-05-07 08:44

本文选题：HOG特征 + 自适应学习率背景建模　；参考：《计算机科学》2017年S1期

【摘要】：行人统计在智能监控领域中具有重要意义,但复杂背景环境以及行人运动过程中出现的遮挡现象导致当前方法的准确率并不高。此外,传统过线统计人数的方式的实际适用范围有限。考虑到现有方法的不足,提出了一种基于热点区域定义的人数统计方法。首先,利用自适应学习率背景建模提取运动目标前景,得到前景区域的位置和大小,扫描计算运动目标前景范围内的HOG特征,并判别是否存在头肩目标;然后,利用基于KCF的目标匹配算法跟踪头肩目标;最后,结合目标运动轨迹与提出的区域人数统计算法进行行人人数统计。采用24fps的手机拍摄的长度为10min、分辨率为960×720像素的视频做人数统计实验。实验结果表明,所提算法在统计人数时正确率可达到93.1%,能满足实时性要求。该方法结合了检测效率和准确率,在背景环境复杂的场景下具有良好的效果,能适应各类人数统计的实际应用场景。
[Abstract]:Pedestrian statistics is of great significance in the field of intelligent monitoring, but the accuracy of the current methods is not high due to the complex background environment and the occlusion phenomenon in the pedestrian movement process. In addition, the traditional method of over-line statistics of the actual scope of application is limited. Considering the shortcomings of the existing methods, a population statistics method based on the definition of hot spot region is proposed. First, using adaptive learning rate background modeling to extract the foreground of moving target, obtain the position and size of foreground region, scan and calculate the HOG feature in the foreground range of moving target, and judge whether there is head-shoulder target. The target matching algorithm based on KCF is used to track the head-shoulder target. Finally, the pedestrian number is calculated by combining the moving track of the target with the proposed regional population statistics algorithm. The 24fps mobile phone is used to record video with a length of 10 minutes and a resolution of 960 脳 720 pixels. The experimental results show that the accuracy of the proposed algorithm can reach 93.1, which can meet the real-time requirements. This method combines the detection efficiency and the accuracy, and has good effect under the background environment complex scene, and can adapt to the practical application scene of all kinds of population statistics.
【作者单位】：浙江工业大学计算机科学与技术学院;
【基金】：国家自然科学基金资助项目(C12412135,61402410) 浙江省自然科学基金资助项目(LY13F020029,LQ14F020004)资助
【分类号】：TN948.6;TP391.41

【相似文献】