Multi - Camera Person of Interest Tracking in Crowded Scenes Using Homography and Re-Identification

In crowded and multi-camera surveillance scenarios, ensuring accurate tracking of individuals across different camera views is a complex challenge. A multi-faceted approach is thus used to address this issue. The YOLOv8 (You Only Look Once) object detection model is used for real-time detection in each camera stream. A Homography Matrix is employed for geometric alignment and calibration of cameras, ensuring a consistent and unified perspective. This technique proves crucial in scenarios where cameras are placed at varying angles and heights. DeepSORT, a state-of-the-art tracking algorithm, utilizes a deep neural network for person re-identification. This algorithm excels in distinguishing individuals, even in densely populated environments. By leveraging the Homography Matrix for geometric alignment and DeepSORT for precise person tracking, the proposed system achieves reliable multi-camera person tracking in challenging and crowded surveillance settings.