Temporal Max Pooling Results in Rainforest Environments

Temporal Max Pooling (specifically only the TMPv channel) was initially used and evaluated during a field expedition in Sumatra’s Way Kambas National Park (TNWK) and Java’s Ujung Kulon National Park (TNUK) during May 3–14, 2026. While the primary objective of the expedition was to confirm the presence of the critically endangered wild Sumatran rhinoceros (Diceorhinus sumatrensis) in TNWK, our method also proved highly effective for detecting various other wildlife species and humans in tropical rainforest ecosystems (Fig. S3 and Supplementary Movies S6-S13). Supplementary Movies: Movie S6. A side-by-side comparison of raw aerial thermal recordings versus TMP results. The imagery captures a semi-wild Sumatran rhino and their keeper at the Sumatran Rhino Sanctuary (SRS) in Way Kambas National Park (TNWK), recorded from an altitude 450–500ft AGL under conditions of canopy occlusion and severe thermal clutter. Original raw video footage (left) and TMP results (right). Video speed-up factors are indicated. The recordings were acquired in daylight from an altitude of 460ft AGL. Movie S7. Screen recording of a walking person being detected and tracked through dense rainforest vegetation. A wide-angle thermal camera was used for initial detection and re-acquisition when the target was lost, while a telezoom camera was deployed for verification. Temporal switching between raw video footage and TMP results. The recordings were acquired in daylight from an altitude of 470ft AGL. Movie S8. Screen recording of detected bird flight trajectories in RGB footage. Temporal switching between raw video footage and TMP results. The recordings were acquired in daylight from an altitude of 540ft AGL. Movie S9. Screen recording of detected perching hornbills in RGB telezoom footage. Original raw video footage (right) and TMP results (left). Video speed-up factors are indicated. The recordings were acquired in daylight from an altitude of 410ft AGL. Movie S10. Screen recording showing a long-tailed macaque detected in thermal footage and verified via telezoom. Temporal switching between raw video footage and TMP results. Video speed-up factors are indicated. The recordings were acquired in daylight from an altitude of 500ft AGL. Movie S11. Screen recording of a black monkey initially detected by conspicuous branch motion in thermal footage and subsequently verified via telezoom. Temporal switching between raw video footage and TMP results. Video speed-up factors are indicated. The recordings were acquired in daylight from an altitude of 390ft AGL. Movie S12. Screen recording of bats hunting insects, displaying the flight trajectories of both. Temporal switching between raw video footage and TMP results. The recordings were acquired at night and from the ground. Movie S13. Screen recording capturing the discovery of a wild Sumatran rhino, with digital zoom highlighting its distinct shape and motion patterns. Temporal switching between raw video footage and TMP results. The recordings were acquired at night from an altitude of 260-500ft AGL. Supplementary Software: Software S3. The real-time processing pipeline of TMP is implemented in Python (3.7+) and includes sample raw video footage from the examples shown in Supplementary Movie S6. To run the script, first copy the desired video file from the SampleRecordings directory into the main folder, then execute TMP.py.The software displays a synchronized, split-screen view showing the original footage on the left and the real-time TMP results on the right. TMP parameters can be interactively adjusted during playback: use the UP/DOWN arrow keys to modify the integration window size N, and the LEFT/RIGHT arrow keys to adjust the frame-skipping interval S.