Deep in the Jungle: Towards Automating Chimpanzee Population Estimation

TL;DR

The paper addresses the challenge of estimating chimpanzee populations from camera traps when individuals are unmarked and direct distances are hard to obtain. It proposes a pipeline that integrates monocular depth estimation (MDE) into camera-trap distance sampling (CTDS) and evaluates four model/configuration combinations on a real-world Taï dataset. Results show that Dense Prediction Transformer (DPT) based pipelines generally yield more accurate distance estimates than Depth Anything, with segmentation improving short-range depth accuracy, though all methods tend to overestimate distances and underestimate density, especially under fully automated processing. This work demonstrates a viable, scalable approach toward automated great-ape monitoring and identifies bottlenecks—calibration, robust localization, and depth-model adaptation to forest environments—that must be addressed for deployment.

Abstract

The estimation of abundance and density in unmarked populations of great apes relies on statistical frameworks that require animal-to-camera distance measurements. In practice, acquiring these distances depends on labour-intensive manual interpretation of animal observations across large camera trap video corpora. This study introduces and evaluates an only sparsely explored alternative: the integration of computer vision-based monocular depth estimation (MDE) pipelines directly into ecological camera trap workflows for great ape conservation. Using a real-world dataset of 220 camera trap videos documenting a wild chimpanzee population, we combine two MDE models, Dense Prediction Transformers and Depth Anything, with multiple distance sampling strategies. These components are used to generate detection distance estimates, from which population density and abundance are inferred. Comparative analysis against manually derived ground-truth distances shows that calibrated DPT consistently outperforms Depth Anything. This advantage is observed in both distance estimation accuracy and downstream density and abundance inference. Nevertheless, both models exhibit systematic biases. We show that, given complex forest environments, they tend to overestimate detection distances and consequently underestimate density and abundance relative to conventional manual approaches. We further find that failures in animal detection across distance ranges are a primary factor limiting estimation accuracy. Overall, this work provides a case study that shows MDE-driven camera trap distance sampling is a viable and practical alternative to manual distance estimation. The proposed approach yields population estimates within 22% of those obtained using traditional methods.

Figures (8)

• Figure 1: Overview of Proposed Approach.(1) Frames containing calibration markers at known distances, and (2) habitat frames sampled from motion-triggered camera trap videos, are (3) passed to MegaDetector and SAM to localise and segment content of interest. These are then processed via (4) monocular depth estimation, utilising the approach of Haucke et al. HAUCKE2022101536, to yield raw depth maps of observations. The framework further combines (5) calibrated metric depth maps with the raw data to produce (6) distance estimates for all detected chimpanzees. These estimates are combined with camera trap metadata (spatial and temporal) and statistically processed to yield (7) density and abundance estimates using the CTDS statistical framework. Overall, this pipeline streamlines camera trap distance sampling from days or weeks of manual labour to hours of computation.
• Figure 2: Chimpanzee Distance Estimation Performance by Distance. Graphs showing both individual (bottom) and mean (top) modelled distance estimates mapped to their corresponding manual estimates for all configurations. The blue dotted line shows the fitted regression line. The red dashed line shows the ideal (i.e., model=manual). The error bars show the 25–75 (green) and 5–95 (purple) percentiles (best viewed zoomed). There is a reasonable correlation between manual and modelled distance, although the level of correlation varies.
• Figure 3: Distance Estimation Error by Ground Truth Distance. Mean average error (left) and root mean squared error (right) for distance estimates grouped by their corresponding manual estimates for each pipeline configuration. MAE is generally high at the extremes of the distance scale with an additional spike at approx. seven meters.
• Figure 4: Close-up Distance Estimations. Example of Depth Anything depth maps generated using bounding box (left) and segmentation (right) detection methods at a detection distance of 0.5 meters (manual estimate). In this example, the bounding box method gives a distance estimate of 4.11 meters while the segmentation method gives a distance estimate of 4.17 meters. The distance estimates are close due to identical depth maps, unlike DPT where calibration is applied.
• Figure 5: Impact of Occlusions on Bounding Box–based Distance Estimation. Example DPT depth maps generated using bounding box–based (top) and segmentation-based (bottom) detection methods at a manual detection distance of 6.5m, where the individual is partially occluded by foliage. The bounding box method estimates a distance of 4.40m, while the segmentation method estimates 6.96m.