---
title: "Build a Multi-Camera 3D Tracking Application with NVIDIA DeepStream 9.1 Skills"
description: "NVIDIA DeepStream 9.1 introduces AutoMagicCalib and Multi-View 3D Tracking skills to resolve the limitations of single-camera 2D tracking and manual camera calibration in large-scale video analytics. Multi-View 3D Tracking projects detections into a shared 3D coordinate system, matching tracklets across cameras via MQTT and maintaining globally consistent object IDs. AutoMagicCalib automates camera intrinsic and extrinsic parameter estimation by analyzing object trajectories across video feeds, optionally using a Visual Geometry Grounded Transformer. These capabilities are integrated into 13 agentic skills that allow coding agents to set up, configure, and execute end-to-end vision AI pipelines from natural language prompts. Outputs include an on-screen display grid, bird's-eye view trajectories, and per-frame protobuf metadata streamed through Kafka."
---

# Build a Multi-Camera 3D Tracking Application with NVIDIA DeepStream 9.1 Skills

[NVIDIA](https://yomu.fyi/company/nvidia-developer-blog) · Elizabeth Goodman · Jul 15, 2026

**Type:** Announcement

## Summary

NVIDIA DeepStream 9.1 introduces AutoMagicCalib and Multi-View 3D Tracking skills to resolve the limitations of single-camera 2D tracking and manual camera calibration in large-scale video analytics. Multi-View 3D Tracking projects detections into a shared 3D coordinate system, matching tracklets across cameras via MQTT and maintaining globally consistent object IDs. AutoMagicCalib automates camera intrinsic and extrinsic parameter estimation by analyzing object trajectories across video feeds, optionally using a Visual Geometry Grounded Transformer. These capabilities are integrated into 13 agentic skills that allow coding agents to set up, configure, and execute end-to-end vision AI pipelines from natural language prompts. Outputs include an on-screen display grid, bird's-eye view trajectories, and per-frame protobuf metadata streamed through Kafka.

## Context

Developers tracking objects across large spaces face limitations with single-camera 2D tracking, which lacks reliable depth information and loses track of objects leaving the frame. Additionally, conventional 3D tracking requires time-consuming manual camera calibration involving physical patterns and operational interruptions.

## Approach / What changed

NVIDIA released DeepStream 9.1 featuring Multi-View 3D Tracking and AutoMagicCalib alongside 13 modular agentic skills. Multi-View 3D Tracking back-projects 2D detections into 3D world-space coordinates via ground-plane projection and associates tracklets across cameras using MQTT. AutoMagicCalib automates calibration by estimating intrinsic and extrinsic parameters from video trajectories through bundle adjustment, microservice REST APIs, and a web UI.

## Takeaways

- Multi-View 3D Tracking uses MQTT to share tracklets across cameras, matching them by 3D world space proximity to maintain a single global ID.
- AutoMagicCalib automates intrinsic and extrinsic parameter calibration from video trajectories and can optionally utilize the Visual Geometry Grounded Transformer model.
- DeepStream 9.1 outputs real-time tracking data via on-screen display grids, bird's-eye view maps, and structured protobuf metadata published to a Kafka topic.

**Tags:** [Architecture](https://yomu.fyi/topic/architecture), [Kafka](https://yomu.fyi/topic/kafka), [Machine Learning](https://yomu.fyi/topic/machine-learning), [Open Source](https://yomu.fyi/topic/open-source), [Streaming](https://yomu.fyi/topic/streaming)

- Source: [NVIDIA](https://developer.nvidia.com/blog/build-a-multi-camera-3d-tracking-application-with-nvidia-deepstream-9-1-skills)
- Source URL: https://developer.nvidia.com/blog/build-a-multi-camera-3d-tracking-application-with-nvidia-deepstream-9-1-skills
- Ingested by Yomu: 2026-08-27T15:01:05.769Z

[Read original post](https://developer.nvidia.com/blog/build-a-multi-camera-3d-tracking-application-with-nvidia-deepstream-9-1-skills)