CPU-GPU heterogeneous implementations of depth-based foreground detection

Abstract

Video sensor data has been widely used in automatic surveillance applications. In this study, we present a method that automatically detects the foreground by using depth information. For real-time implementation, we propose a means of reducing the execution time by applying parallel processing techniques. In general, most parallel processing techniques have been used to parallelize each specific task efficiently. In this study, we consider a practical method to parallelize an entire system consisting of several tasks (i.e., low-level and intermediate-level computer vision tasks with different computational characteristics) by balancing the total workload between CPU and GPU. Experimental results with a pig monitoring application reveal that the proposed method can automatically detect the foreground using CPU-GPU heterogeneous computing platforms in real time, regardless of the relative performance between the CPU and GPU.

A novel NVM memory file system for edge intelligence

“…It can significantly provide additional computing power and reduce data movement. Most heterogeneous NDC systems are based on GPU compute units, but the naturally high power consumption of GPUs makes them unsuitable for edge deployment [3,4,5,6,7]. FPGA becomes candidate as co-processing units in edge NDC due to its parallel computing power and low power consumption, and its flexibility can adapt to the emerging AI algorithms.…”