红外实时人体目标检测因无需主动光源、易于去除环境干扰等优点成为近年来人工智能领域的一个研究热点。当前主流实时红外嵌入式图像采集处理方式因图像传输、接口速度、工程量等技术局限, 制约了红外人体目标检测的便捷实现和实时利用。针对传统嵌入式处理器难以同时采集和处理红外数据这一影响红外人体目标检测实时性的技术问题, 研究了基于ZYNQ异构多核处理器的实时检测技术。采用FPGA+多核ARM架构, 由FPGA端实现人体红外图像的VDMA采集、预处理及显示, 由多核ARM端完成HOGSVM算法处理, 并使用TBB进行加速。实验结果表明: 该技术能够完成人体目标的实时有效检测, 兼具功耗低、开发简单、实用性强和集成度高等优点。

The realtime infrared human target detection has become a research hotspot in the field of artificial intelligence in recent years because of its advantages of no need of active light and easy removal of environmental interference. The current mainstream realtime infrared embedded image acquisition and processing methods have limited the convenient realization and realtime utilization of infrared human body target detection due to the technical limitations of image transmission, interface speed and engineering quantity. In this paper, realtime detection technology based on ZYNQ heterogeneous multicore processor is proposed in view of that the traditional embedded processor is hard to collect and process infrared data at the same time, which affects the realtime detection of infrared human target detection. The technology uses FPGA+ multicore ARM architecture, FPGA realizes the human body infrared image VDMA acquisition, pretreatment, display, multicore ARM realizes human HOGSVM algorithm and using TBB to accelerate. Experimental results show that the technology can accomplish the realtime and effective detection of human target with the advantages of low power consumption, simple development and strong practicability.