Deep learning-based monitoring and spatiotemporal analysis of wildlife and human activities using camera trap imagery

Abstract Wildlife and human activities are key components of ecological systems, and understanding their spatiotemporal patterns is essential for evaluating human–wildlife interactions and informing conservation planning. Camera trap networks function as terrestrial remote sensing systems that enable continuous monitoring of ecological processes at fine spatiotemporal scales. In this study, camera trap imagery collected in Chitwan National Park (CNP), Nepal, and its surrounding zones, under visible light during daytime and infrared illumination at night, was organized into a multi-class dataset containing wildlife, humans, and domestic animals. To support the ecological interpretation of wildlife–human spatiotemporal patterns, deep learning–based object detection models (YOLOv5 and YOLOv11) were trained to extract animal and human occurrences from the imagery. The resulting detections were integrated into a spatiotemporal analysis framework to characterize wildlife and human activity patterns, map spatial co-occurrence across the landscape, and quantify temporal proximity between human-wildlife pairs. Among the evaluated models, YOLOv11s achieved the best detection performance, with a precision of 95.5%, recall of 92.9%, and mean Average Precision50 (mAP50) of 95.4%, indicating reliable performance in real-world scenarios. Spatiotemporal analysis revealed shorter temporal lags for human–rhino pairs compared to human–tiger pairs, and identified a clear spatial gradient of human–wildlife co-occurrence, with the highest intensity in the Buffer Zone, followed by the Corridor Forest and the CNP. These results demonstrate that integrating automated detection with spatiotemporal analysis transforms camera trap data into ecological information for scalable monitoring. The proposed framework identifies areas of elevated human–wildlife spatiotemporal proximity, helping prioritize targeted monitoring and landscape management.