如何使用 Flutter 優化 NVIDIA Jetson 上的相機與影片管線

Optimizing Camera and Video Pipelines on NVIDIA Jetson with Flutter

Focus: Jetson camera capture, GStreamer, hardware acceleration,
Flutter preview\
Audience: Flutter, Android, edge AI, and robotics engineers

The problem

Video pipelines can waste Jetson CPU and memory when frames are
repeatedly copied between camera, decoder, inference, network, and UI
layers. A Flutter preview should not force the robot's perception
pipeline to run at display rate.

What this guide covers

This tutorial focuses on practical architecture and performance
engineering for production systems. The central principle is to measure
the complete pipeline, keep latency-sensitive work close to the
hardware, and make failure states visible to the operator.

Reference architecture

Camera → GStreamer/V4L2 capture → hardware-accelerated decode/convert
where supported → inference branch and preview branch →
encoded/downsized stream or sampled frames → Flutter UI.

Implementation plan

Profile capture, decode, color conversion, inference, encoding, and network transfer independently.
Use hardware-accelerated GStreamer elements where supported by the Jetson model and software image.
Keep inference and preview branches decoupled; preview may sample frames without throttling perception.
Prefer encoded transport for remote preview; avoid sending raw RGB frames unless resolution and bandwidth justify it.
Track dropped frames, queue depth, GPU/CPU load, memory bandwidth, and thermal throttling.

Example

# Illustrative GStreamer pipeline shape; element names and availability\n# depend on JetPack, camera source, codec, and Jetson generation.\nsource ! queue leaky=downstream max-size-buffers=2 !\nconvert_or_scale ! encoder_or_network_sink

The snippets are illustrative starting points, not drop-in integrations.
Match APIs and version requirements to the exact SDK, OS image, hardware
revision, and runtime in your project.

Performance checklist

Measure end-to-end latency, not only one function or model invocation.
Report p50 and p95/p99 latency alongside throughput, memory use, and dropped work.
Bound queues and define what happens when consumers fall behind.
Avoid repeated frame copies and expensive work on Flutter's UI isolate.
Test sustained workloads after the device reaches thermal equilibrium.
Keep UI state, device state, and robot state distinct.
Add timeouts, cancellation, reconnection, and explicit stale-data indicators.
Profile on the actual target hardware using a reproducible build.

Common pitfalls

Do not copy a pipeline from another JetPack release without checking installed plugins.
Avoid unbounded queues: they turn overload into increasing latency.
Do not assume that hardware encoding is available for every codec/resolution combination.