Distributed FPGA-based smart camera architecture for computer vision applications

Abstract

International audienceSmart camera networks (SCN) raise challenging issues in many fields of research, including vision processing, communication protocols, distributed algorithms or power management. Furthermore, application logic in SCN is not centralized but spread among network nodes meaning that each node must have to process images to extract significant features, and aggregate data to understand the surrounding environment. In this context, smart camera have first embedded general purpose processor (GPP) for image processing. Since image resolution increases, GPPs have reached their limit to maintain real-time processing constraint. More recently, FPGA-based platforms have been studied for their massive parallelism capabilities. This paper present our new FPGA-based smart camera platform supporting cooperation between nodes and run-time updatable image processing. The architecture is based on a full reconfigurable pipeline driven by a softcore

SPINE: From C loop-nests to highly efficient accelerators using Algorithmic Species

“…Each accelerator receives data via an AXI streaming interface and stores this in a local buffer and sends data from its output buffer via another AXI interface. The same is done in the (AS) 2 and HLS implementations that we will use as a reference, but these do not include the actual AXI interface while SPINE does. The results presented are post place-and-route results and resource usage (U ) is calculated as follows:…”

“…Examples of such systems can be found in industrial applications such as search engine acceleration [1] and smart cameras [2], and consumer applications such as 4G communication in smartphones and wearable devices.…”

An FPGA‐based smart camera for accurate chlorophyll estimations

“…Smart cameras are image/video acquisition devices with self‐contained image processing algorithms that simplify the formulation of a particular application . For example, algorithms for smart video surveillance could detect and track pedestrians, but for a robotic application, algorithms could be feature detection or feature tracking .…”

Robust feature extraction algorithm suitable for real-time embedded applications

“…For instance, algorithms for smart video surveillance could detect and track pedestrians [23], but for a robotic application, algorithms could be edge and feature detection [2]. In recent years, advances in embedded vision systems such as progress in microprocessor power and FPGA technology allowed the creation of compact smart cameras with low cost and, this increased the smart camera applications performance, as shown in [11,6,7,8]. In current embedded vision applications, smart cameras represent a promising on-board solution under different application domains: motion detection [25], object detection/tracking [32,31], inspection and surveillance [16], human behavior recognition [20], etc.…”