Build a Multi-Camera 3D Tracking Application with NVIDIA DeepStream 9.1 Skills | NVIDIA Technical Blog

DeepStream 9.1 introduces AutoMagicCalib (AMC) and Multi-View 3D Tracking (MV3DT) skills to automate camera calibration and enable consistent object tracking across multiple camera views. AMC automates the estimation of intrinsic and extrinsic camera parameters using DeepStream's object tracking data and optional VGGT integration. MV3DT fuses detections from multiple calibrated cameras into a shared 3D coordinate system, assigning globally consistent object IDs. The system supports PeopleNetTransformer, PeopleNet v2.6.3, and RT-DETR 2D detectors, and outputs results via OSD, BEV, and Kafka. The post also highlights the integration of agentic skills for natural language-driven pipeline deployment.