结合多特征融合算法的人数统计与分析系统

People Counting System Combined with Multi-Feature Fusion Algorithm

摘要: 为了实现高准确率的实时人数统计，设计了嵌入式片上系统平台。该系统采用交叉编译的方式，对人体目标进行实时检测与跟踪，实现了对指定区域内的人数统计与分析。本系统利用了AdaBoost分类器对图像/视频中目标进行人脸识别，基于Haar特征提取进行目标跟踪。提出结合多特征融合算法以及人脸识别参数自适应算法，在Xilinx公司的Zynq-7000开发板上实现了目标实时性和快速性的统计与分析。结果表明：该系统充分节省了嵌入式平台的资源，简化了整个系统的开发流程，并提高了系统的兼容性和可移植性。结合多特征的融合算法使人数统计的准确率高达97%以上。该系统具有安装位置灵活，实时性好、准确率高、稳定性好等特点。

Abstract: In order to achieve high accuracy people counting, the embedded platform is designed in this pa-per. The system realizes real-time target detection and tracking using the cross compiler, to achieve the number of people within the designated area. The system uses the AdaBoost classifier image/video target recognition and feature extraction based on Haar for target tracking. The combination of multi-feature fusion algorithm and face recognition parameters adaptive algorithms is proposed in Xilinx’s Zynq-7000 development board to achieve the goal of real-time and fast statistics and analysis. The results show that: The system saves resources of embedded platforms and simplifies the entire system development process, improving the whole compatibility and portability. It can realize people counting and analysis in the image/video with accuracy rate of more than 97%. Basically it meets the requirements of human counting in traffic statistics: Parallel processing, high accuracy, stable and reliable, etc.