Real-time multiple video streams processing on PC-based FPGA platform

Programmable embedded media server based video surveillance system is prevalent as it has the capability to accommodate large network of cameras. The implementation of a wide range of programmable security analytics application in PC-based platform demands high processing bandwidth and high-end hardware specification requirements. This is due to insufficient compute resource by multicore CPU in performing some of the computationally intensive analytics operations at acceptable data rates. This paper presents a platform design considerations for programmable video analytics solving compute intensive video data processing leading to cost-efficient embedded platform solution. Our experimental analysis covers PC-based processing pipeline analysis for CPU resources to be used as design input for processing data offloading between multicore CPU and FPGA. The result identifies the processing bottleneck in PC-based media server with video analytics processing component consumes 81% of the total CPU processing cycles. This leads to the exploration of the PCI Express capability in FPGA for system solution as a transfer medium between these computing units.