Skip to content

README.md

Latest commit

 

History

History
138 lines (92 loc) · 4.76 KB

README.md

File metadata and controls

138 lines (92 loc) · 4.76 KB

4.9 DeepStream and Jetson

Why This Matters

By this point in the course, the learner understands computer vision tasks, model training, export, and real-time pipelines. The next step is to see how those ideas scale into a practical edge AI system.

This is where DeepStream and Jetson become important.

DeepStream is not just another demo tool. It is a framework for building high-performance streaming analytics systems. On Jetson devices, it helps connect camera input, inference, tracking, metadata, and system services into a more production-like architecture.

Learning Objectives

By the end of this section, you should be able to:

  • explain what DeepStream is and why it matters
  • understand why Jetson is a strong edge AI platform
  • describe the roles of JPS, VST, Redis, Ingress, and Analytics
  • understand how multi-stream systems differ from single-stream demos
  • run and inspect a basic DeepStream/JPS-oriented workflow

Core Concepts / Theory

Why Jetson for Edge Vision

Jetson devices are designed for edge AI workloads where local inference matters. Common reasons to use Jetson include:

  • low-latency local processing
  • reduced cloud dependency
  • GPU acceleration on embedded hardware
  • support for real-time video analytics

What DeepStream Adds

Compared with a simple script that reads a camera and runs a model, DeepStream provides:

  • hardware-accelerated decode
  • batch-friendly processing
  • integrated tracking
  • metadata generation
  • better support for multi-stream analytics

What Jetson Platform Services Adds

Jetson Platform Services adds a service-oriented layer around the pipeline.

Important components include:

  • VST for stream onboarding and video handling
  • Redis for metadata and service communication
  • Ingress for unified external access
  • Analytics for rules such as ROI and line crossing

From Demo to System

A single demo script often answers:

  • can I read one camera
  • can I run one model

A system-level deployment must answer more:

  • can I scale to multiple streams
  • can I manage metadata
  • can I recover from instability
  • can I expose services to operators and other applications

That is the system mindset this section introduces.

Key Terms

  • DeepStream: NVIDIA's streaming analytics framework
  • JPS: Jetson Platform Services
  • VST: Video Storage Toolkit
  • Redis: in-memory data store often used as a metadata backbone
  • Ingress: routing layer for external service access
  • Analytics: service for rule-based event logic

Worked Example / Code Example

Start Core Services

sudo systemctl start jetson-redis
sudo systemctl start jetson-ingress
sudo systemctl start jetson-vst

Check That JPS Services Exist

ls /opt/nvidia/jetson/services
systemctl status jetson-redis
systemctl status jetson-vst

Typical DeepStream Sample Run

deepstream-app -c /opt/nvidia/deepstream/deepstream/samples/configs/deepstream-app/source30_1080p_dec_infer-resnet_tiled_display_int8.txt

Inspect a Managed Stream Path

gst-launch-1.0 rtspsrc location=rtsp://<jetson-ip>:8554/<stream> latency=200 ! \
  rtph264depay ! h264parse ! avdec_h264 ! videoconvert ! autovideosink

Common Misunderstandings

  • "DeepStream is only for benchmarking."
    • It is also a system-building framework for streaming analytics.
  • "If one camera works, multi-stream deployment is just more of the same."
    • Multi-stream systems introduce scheduling, metadata, and stability concerns.
  • "Jetson deployment is only about converting the model."
    • Deployment also includes input management, services, monitoring, and event logic.

Exercises / Reflection

  1. Explain in your own words how DeepStream differs from a simple OpenCV inference script.
  2. List three reasons a multi-stream deployment is harder than a single-stream demo.
  3. Draw a simple block diagram showing how camera -> DeepStream -> tracking -> metadata -> analytics might flow.
  4. Reflect on when a learner should choose Jetson local deployment instead of cloud-only processing.

Summary

DeepStream and Jetson turn computer vision into a full edge system. This section helps the learner shift from thinking about one model on one stream to thinking about scalable, service-based visual analytics.

Suggested Next Step

Continue to 4.10 Frontier Vision Technologies and Outlook, then explore the AI NVR on reComputer appendix.

References