Deep SORT brand tracking in video

Detection alone was not enough for video: sponsors care about exposure over time. I orchestrated a containerized Deep SORT tracking solution that followed detected brands across frames and produced real-time exposure insights.

The system scaled to 1M+ video frames per week with sub-200ms inference latency using GPU acceleration, so analytics could run continuously on incoming social video rather than as an offline batch job.

Focus

• Deep SORT multi-object tracking for brand exposure
• Containerized deployment with GPU acceleration
• Real-time video analytics at million-frame scale