Automated Detection and Visualization of Meetingroom Attendance

Sammanfattning

Automated analysis of human presence and movement is important in areas such as smart buildings and indoor occupancy monitoring. Accurate detection and localization of humans can help improve space utilization and safety. A main challenge in this project is achieving reliable real time human detection and spatial position estimation, while reducing noise and false detections. In this thesis, the objective is to design and implement a system for real time human detection, three dimensional position estimation, and visualization using an embedded vision platform. The proposed approach combines a stereo camera setup with the YOLOv6 object detection model. Human detection are filtered based on confidence thresholds, class selection, and non-maximum suppression. Spatial coordinates are computed using depth information, and temporal averaging is applied to improve stability. The positions are visualized in a two dimensional top down view. The results show that the system reliably detects multiple humans and produce stable position estimate.