DeepStream Reference Designs/Getting Started/Environment Setup

DeepStream Reference Designs
Getting Started
Environment Setup Evaluating the Project Purchasing a License
Project Architecture
High Level Design
Reference Designs
Automatic Parking Lot Vehicle Registration
Customizing the Project
Implementing Custom Inference Parser Implementing Custom Inference Listener Implementing Custom Policies Implementing Custom Actions Implementing Custom Media Implementing a Custom Application
Contact Us

This page describes the fundamentals of RidgeRun's DeepStream Reference Designs. After reading this page it should be clear if the product meets your requirements for building your own applications.

Introduction

RidgeRun's DeepStream Reference Designs is a project that provides a robust and modular design, based on the NVIDIA DeepStream SDK, where the building blocks may be replaced to fit a wide variety of use cases. The main objective is to provide an infrastructure for an application using video analytics to perform informed decisions within the application domain. The system could be divided into the following parts:

Framework

The framework is the main infrastructure responsible for driving the application state and logic. All modules here do not need modification in order to implement a new application, consequently, this part is the common source that is shared with other applications. There are four principal sections that compose the framework:

• Camera Capture: In charge of keeping the control of media sources (cameras, RTSP streams).
• AI Manager: This module will process DeepStream inference defined by the application and will communicate with the next section.
• Action Dispatcher: Uses the data of DeepStream inference to do established actions depending on policies. The actions and policies are defined by the application.
• Config Parser: This module is in charge of loading the configuration files set up to be used by the application.

Application

We focus on bringing a solution that the users only need to think about what are they looking for and not develop everything right from the start. So in this part, the users can define the camera source, DeepStream model, and policies that do they serve as a filter of DeepStream inference data that allow for implementation decisions and be executed by the actions, which are also user-defined. All the logic and mechanism to execute all that the user uses in the application is provided by the framework.

System Setup

We offer several versions of the system, all based on the NVIDIA Jetson boards. Small places that use a single or dual camera may use a Jetson Nano, while big commerce with 32 cameras may need to use the Jetson Xavier AGX.

• RTSP Cameras: 1 to 8 simultaneous cameras.
• Jetson Board: portable through Jetson Nano, TX2, NX, and AGX development kits.

The system is based on NVIDIA's DeepStream to take advantage of the board's hardware and multimedia capabilities.