Accelerate Smart City Applications

Xilinx accelerates Huati Roadside Unit (RSU) for Smart Roadside Parking and V2X applications

 

Introduction

Huati is one of the leading companies of the smart street light and other smart city application systems in China.

As reported by EEWorld.com.cn, according to the analysis of the Industry Research Center of China, it is estimated that the size of China’s smart city market reached 6 trillion yuan in 2017, and the average annual compound growth rate in the next five years (2017-2021) will be approximately 32.64%. As the key component of smart cities, the scale of the global smart street light market is also very promising. According to the Industrial Research Institute, the global smart street light market will exceed US $6 billion by 2020 and due to urbanization and the development of smart cities, the Chinese market is expected to exceed 10 billion yuan by 2020.

Roadside Unit (RSU) is one of the key devices in the smart street light system to enable the emerging smart roadside parking and V2X applications. The complexity and ever evolving nature of modern cities and AI algorithms make it necessary for the AI based smart city devices such as to have adequate computation power, low latency, to be upgradable and low cost.

Huati’s CloudEye RSU is designed to meet all these goals, thanks to the excellent balance of cost, performance and flexibility of the industry-leading FPGA platforms from Xilinx.

Together with smart street light infrastructures, the deployment of smart city devices has become unprecedentedly fast, efficient and cost effective.

 

Key Advantages

Compare to the similar embedded GPU solutions, the FPGA based RSU solution has the following advantages:

FPS 4 x
Computational Efficiency (FPS/GOPS) 4 x
Energy Efficiency (FPS/Watt) 2.6 x
Latency 1/70

 

Solution Overview

The roadside smart parking system is the cutting-edge computer vision based solution. With advanced AI algorithms, the system intelligently identifies the complete parking trajectory, with 99% comprehensive recognition output rate. Combined with the electronic payment system, it perfectly achieves the goals of no parking attendant and unconscious pay experience.

The system is also integrated with automated driving and driver assistance systems, to provide V2X features such as Automated Valet Parking(AVP), traffic avoidance, traffic data collection and analysis, etc.

The smart street light was introduced in recent years as a new entrance to smart city solutions.

More and more smart city devices, such as 4G/5G base stations, RSUs for smart parking and V2X applications, EV charging stations, digital
billboards, surveillance cameras, environment monitoring, etc. have been integrated into the smart street lights.

During the construction, optical fiber high speed communication and large capacity power supply infrastructures are also put in place
together with the smart street light network to get ready for future smart city solutions.


CloudEye RSU

Huati RSU, CloudEye, are the neurons of the smart street light infrastructures.

The edge server provides computer vision and deep learning engines for applications such as smart road side parking, smart city object recognition, identification and data analyzation.

It also communicates to vehicles equipped with automated driving and driver assistance systems to provide information to support V2X features such as AVP, traffic avoidance and traffic data collection and analysis, etc.

 

HT-AIoT Module

CloudEye RSU is powered by the Xilinx Kintex-7 FPGA based AIoT module, which is designed by Nunova. With deeply customized artificial intelligence networks, this AIoT core solution is the perfect combination of high computing power, low power consumption and low latency for computer vision applications.

Key Features:

AI chip hardware algorithm acceleration, multiple deep learning network tasks in parallel processing, automated machine learning mechanism

Ultra-high average object recognition accuracy (mAP) and recognition rate

Supports simultaneous processing of up to 8 HD camera video streams

Core board clocked up to 2.0GHz, equipped with PCIe Gen2, RJ45, USB3.0 and other high-speed interfaces

4G DDR3 provides large storage space

 

Future Work

The next-generation of CloudEye RSU will be based on Xilinx ZU+ EV platform which integrates quad Cortex-A53 CPU, 4K@60fps H.264/H.265 VCU and rich programmable logic resource in single-chip SoC which significantly improves the system performance and reduce the PCB footprint and BOM cost.

One ZU+ EV device is capable of handling 8x H.264/H.265 1080p@30fps streams and implementing the real-time AI analytics to extract the necessary metadata using Xilinx DPU in programmable logic. Also, the pre and post-processing including color space conversion, scaling, Softmax etc. can be accelerated with programmable logic to enable end-to-end application acceleration.

The multiple heavy-duty AI tasks including car plate recognition, vehicle identification, pedestrian detection, traffic flow monitoring etc. can be deployed on the DPU built-in hardware platforms through Xilinx Vitis AI.

The stack of Vitis AI is shown below. The multi-channel multi-task AI applications can run efficiently on multiple DPUs with multi-threading and QoS-based task scheduling enabled by the underlying AI Runtime and Library. High-level C++ and Python API as well as Gstreamer are supported for application development.

For more information about Huati, please visit https://www.huati-en.com/

For more information about ZU+ EV, please visit https://www.xilinx.com/products/silicon-devices/soc/zynq-ultrascale-mpsoc.html

For more information about Vitis AI, please visit https://github.com/Xilinx/Vitis-AI

Categories: Technology