Wildlife monitoring is crucial for informing conservation efforts, providing valuable insights into use of wildlife infrastructure, movement patterns, and population dynamics. Traditional monitoring methods that include motion activated cameras and manual analysis are time-consuming and costly. Large-scale wildlife monitoring projects, such as the I-90 Snoqualmie Pass East Project studied here, involve extensive networks of motion-activated thermal cameras that generate millions of images and videos annually. We propose and demonstrate a novel solution: a morphing workflow to convert large datasets of annotated optical imagery to equivalently-annotated thermal-analog imagery. This synthetic thermal data enables training of a high-precision detection model without access to real thermal data. The detection model achieved over 99% accuracy across three animal groups (deer/elk, bobcat, and coyote) and successfully filtered 98.63% of false positives in real-world thermal videos collected at the 61.5 Overcrossing South on the I-90 corridor. The filtered detections were passed to a lightweight classification model trained on real thermal imagery, which achieved species-level classification accuracies of 93% (bobcat and coyote) and 97% (deer/elk). This two-stage system substantially reduces manual review while maintaining high identification performance. By combining scalable synthetic data with targeted real-world training, this approach removes key barriers to large-scale, automated wildlife monitoring using thermal sensors.