Purpose: This study investigates the application of deep learning based object detection to construct a motion heatmap of visitor crowds at Elephant Park, Bandar Lampung, as a monitoring measure during the COVID-19 pandemic. The research aims to determine whether an object detection algorithm can identify crowd density patterns and to evaluate the performance of the detector under varying confidence thresholds.
Research Methodology: The Single Shot MultiBox Detector with a MobileNet V1 backbone, pretrained on the Common Objects in Context dataset, was deployed on an NVIDIA Jetson Nano connected to a USB camera positioned at two intersections inside the park. Background subtraction in OpenCV was combined with detected bounding boxes to generate a motion heatmap. Detector performance was evaluated using a confusion matrix across confidence threshold values ranging from 0 to 0.9.
Results: The detector achieved a mean average precision of 23.2%, with an accuracy of 0.60, precision of 0.84, recall of 0.63, and F1 score of 0.72 at a confidence threshold of 0.3. A confidence threshold of 0.6 produced the most balanced trade-off for generating a usable motion heatmap, yielding a precision of 0.87.
Conclusions: The combined detection and heatmap pipeline successfully visualized visitor flow patterns at both observation points.
Limitations: Detection accuracy degraded under intense lighting variation and small pedestrian scale.
Contributions: This work offers a low cost, embedded surveillance approach that can inform public space design and health protocol enforcement in open air tourist attractions.