From Hosted Service to Field-Owned Infrastructure: Open, Self-Hosted Real-Time Camera-Trap Alerting with Regional Open-Weight Models
Organizations: Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, United Kingdom
Abstract
Cellular camera traps can alert field teams within minutes, supporting anti-poaching, threatened-species protection and human-wildlife conflict mitigation. Most systems use central servers, sending images of people and sensitive species locations outside the collecting organisation and often across national borders. We ask whether this capability can run on infrastructure owned by field organisations. Trap Tracker RT is an open-source, self-hosted application that receives images by email or locally via FTP/SFTP, classifies them with a project-specific model and sends alerts through a mail server chosen by the organisation. We evaluated DeepFaune-SSA v1.2, pairing MegaDetector v6 with a DINOv3-based classifier for 33 sub-Saharan taxa, against three open alternatives on 31,131 out-of-distribution images from Kenya, Namibia and Malawi, independent of its training data. DeepFaune-SSA achieved image-level species accuracy of 0.901 (95% CI 0.894-0.908), versus 0.530 for the global SpeciesNet classifier and 0.556 and 0.547 for YOLOv10x and YOLO26x detectors trained on the same regional data. DeepFaune-SSA had the highest recall for all 26 species evaluated. The single-stage detectors missed animals in about a quarter of images; SpeciesNet assigned species labels in only 57%, usually falling back to coarser labels otherwise. In bench tests of the fully local path, alerts were issued about 6 s after upload completion, including with slowed and interrupted uploads. The two-stage pipeline processed 19 images per minute on eight CPU threads without a graphics processor. Trap Tracker RT is used by five organisations in the United Kingdom and Kenya, including for human-elephant conflict mitigation. We discuss how data paths, rather than inference location, determine data sovereignty, the responsibilities of self-hosting and the ethics of real-time monitoring.
Figures & tables
| System | Where inference runs | How images arrive | Real-time |
|---|---|---|---|
| Conservation AI (retired) [ 18 ] | Hosted servers, United Kingdom | Email from 3G/4G cameras | Yes |
| Gabon smart camera [ 14 ] | Edge (Raspberry Pi bridge) | Label sent over Iridium satellite | Yes, median 7.35 min |
| TrailGuard AI [ 61 , 15 ] | Edge (on-camera processor) | Cellular, radio or satellite | Yes, about 30 s (tigers and people) |
| Instant Detect 2.0 [ 62 ] | Base station and cloud | LoRa, then Iridium | Yes |
| Sentinel [ 63 ] | Edge retrofit | Satellite or cellular | Near real-time |
| Wildlife Insights [ 56 ] | Hosted (cloud) | Batch upload | No |
| Country | Law | Transfer provision |
|---|---|---|
| South Africa | Protection of Personal Information Act 2013 [ 70 ] | s72: transfer abroad only with comparable protection, consent, or listed exceptions |
| Kenya | Data Protection Act 2019 [ 71 ] | ss48–50: appropriate safeguards proven to the Data Commissioner; s50 allows processing to be required in Kenya |
| Uganda | Data Protection and Privacy Act 2019 [ 72 ] | s19: equivalent protection or consent; s1(b) extends the Act to processors abroad |
| United Kingdom | UK GDPR [ 73 ] | Chapter V (Articles 44A–50, as amended by the Data (Use and Access) Act 2025) |
| Model | Architecture | Training data | Licence |
|---|---|---|---|
| DeepFaune-SSA v1.2 [ 39 ] | MegaDetector v6 + DINOv3 ViT-L classifier | 52,292 regional images, 33 taxa | CC BY-NC 4.0 (weights) |
| SpeciesNet 4.0.3a [ 30 ] | MegaDetector v5a + EfficientNetV2-M classifier, geofenced | More than 65 million images, more than 2,000 labels | Apache 2.0 |
| YOLOv10x v1.0.0 [ 31 ] | Single-stage YOLOv10x, 640 px | Same regional data, 34 taxa plus person and vehicle | CC BY-NC 4.0 |
| YOLO26x v1.0.0 [ 31 ] | Single-stage YOLO26x, 640 px | As YOLOv10x | CC BY-NC 4.0 |
| Dataset | Country and habitat | Images used | SSA species |
|---|---|---|---|
| Biome Health Project Maasai Mara 2018 [ 89 ] | Kenya, savanna | 10,803 | 22 |
| Desert Lion Conservation [ 90 ] | Namibia, desert | 9,272 | 13 |
| Nkhotakota Camera Traps [ 91 , 92 ] | Malawi, miombo woodland | 11,056 | 16 |
| Subset | Images | Accuracy (95% CI) | Macro-F1 (95% CI) | Animal found |
|---|---|---|---|---|
| All, exact species | 27,678 | 0.901 (0.894–0.908) | 0.902 (0.886–0.910) | 0.984 |
| All, genus level | 31,131 | 0.903 (0.896–0.909) | 0.903 (0.888–0.911) | 0.984 |
| Desert Lion, exact | 8,728 | 0.913 (0.905–0.920) | 0.882 (0.863–0.939) | 0.982 |
| Maasai Mara, exact | 9,894 | 0.894 (0.884–0.905) | 0.922 (0.900–0.931) | 0.980 |
| Nkhotakota, exact | 9,056 | 0.897 (0.880–0.911) | 0.922 (0.901–0.933) | 0.991 |
| Subset | DeepFaune-SSA v1.2 | SpeciesNet | YOLOv10x | YOLO26x |
|---|---|---|---|---|
| All | 0.901 (0.894–0.908) | 0.530 (0.503–0.557) | 0.556 (0.539–0.574) | 0.547 (0.529–0.566) |
| Desert Lion | 0.913 (0.905–0.920) | 0.630 (0.617–0.645) | 0.643 (0.629–0.658) | 0.619 (0.604–0.635) |
| Maasai Mara | 0.894 (0.884–0.905) | 0.614 (0.586–0.640) | 0.542 (0.514–0.569) | 0.555 (0.527–0.585) |
| Nkhotakota | 0.897 (0.880–0.911) | 0.342 (0.303–0.381) | 0.488 (0.449–0.525) | 0.469 (0.433–0.502) |
| Animal found (all) | 0.984 | 0.977 | 0.766 | 0.755 |
| Macro-F1 (all) | 0.902 | 0.606 | 0.617 | 0.613 |
| Scenario | Upload | Result | Alert after upload finished |
|---|---|---|---|
| First image after start | 0.35 s | Giraffe, 0.92 | 70 s (GPU warm-up) |
| Slow upload, weak 4G | 40 s, steady | Zebra, 0.93 | 6 s |
| Stalled upload | 20 s pause halfway | Wildebeest, 0.94 | 6 s |
| Slow upload, before the readiness check | 40 s, steady | Read 1 s in: no animal found, no alert | Missed |
| Hardware | Model | Mean | 95th percentile | Images/min | First image |
|---|---|---|---|---|---|
| GPU (RTX A6000) | Two-stage DeepFaune-SSA | 0.11 s | 0.15 s | 542 | 58 s |
| CPU, 32 threads | Two-stage DeepFaune-SSA | 1.29 s | 1.83 s | 47 | 1.4 s |
| CPU, 8 threads | Two-stage DeepFaune-SSA | 3.18 s | 4.69 s | 19 | 3.2 s |
| CPU, 32 threads | Single-stage YOLOv10x | 0.27 s | 0.29 s | 225 | 0.3 s |
| CPU, 8 threads | Single-stage YOLOv10x | 0.72 s | 0.89 s | 84 | 0.8 s |
| GPU (RTX A6000) | Single-stage YOLO26x | 0.08 s | 0.12 s | 744 | 30 s |
| Country | Site and partners | Target | Scale and period | Outcome reported |
|---|---|---|---|---|
| South Africa | Welgevonden Game Reserve, Limpopo; Gap Africa Projects; Vodacom [ 16 ] | Rhinoceros and other large mammals | 27 Reolink 4G cameras, May 2022 to Feb 2023; 37,900 images (18,520 with detections) | 12 species detected; weak 3G/4G signal; cameras destroyed by elephants |
| Uganda | Murchison Falls National Park; Chester Zoo, Uganda Wildlife Authority [ 95 ] | Giant pangolin | About 10 Reolink 4G cameras; project since 2018, Murchison Falls since 2022 | Email alert within 4 minutes of a pangolin detection |
| Kenya | Mount Kenya Wildlife Conservancy; Chester Zoo, Kenya Wildlife Service [ 96 ] | Mountain bongo | About 40 cameras from April 2024 | Alerts used by rangers to locate and tag newborn calves |
| UK | 11 curlew nests, Wales; GWCT Wales; Reolink; Vodafone [ 17 ] | Curlew and chicks | 1,072 images analysed | Accuracy 91.2%; curlew F1 0.951, chick F1 0.960 |
| UK | Knowsley Safari [ 18 ] | Paddock animals | Early deployment, still collecting in 2024 | Not reported |
| Country | Organisation | Since | Cameras | Model | Purpose |
|---|---|---|---|---|---|
| UK | Animal and Plant Health Agency (APHA), Defra | 2024 | About 10 | YOLOv10x, then YOLO26x [ 37 ] , UK mammals | Monitoring grey squirrels |
| UK | Fieldmark AI, Maes Truan Farm, Wales | Jul 2026 | 13 Reolink 4G | DeepFaune-UK, two-stage [ 36 ] | Farm biodiversity study; close to 50,000 images processed |
| UK | GWCT Wales | Early 2026 | About 60 Reolink 4G | YOLO26x, UK mammals [ 37 ] ; previously YOLOv10x | Monitoring farm gates and curlew nests |
| Kenya | Friends of Karura Community Forest Association, Nairobi | Feb 2026 | About 10 Reolink 4G | YOLO26x, sub-Saharan Africa | Wildlife monitoring |
| Kenya | Project PARC | Sep 2026 | 3 Reolink 4G | YOLO26x, sub-Saharan Africa | Human–elephant conflict: an elephant alert e-mail opens a valve on a water supply to a shared community well |
| Configuration | Images pass through | Alerts pass through | Set-up effort | Suited to |
|---|---|---|---|---|
| Hosted service (e.g. Conservation AI, now retired) | Mail provider, then a foreign server | Foreign server, mail provider | Lowest | Sites with no local computer and no sensitive data |
| Trap Tracker RT with a cloud mailbox | Mail provider only | Mail provider | Low: a mail account | Sites where people rarely appear and species are not targeted |
| Trap Tracker RT, fully local | Nothing outside the site | Local relay; its onward route is the site’s choice | Higher: FTP server, relay, network route for cameras | Protected species, people in frame, laws requiring local processing |
Appendix figures & tables5 assets
Supplementary material from the paper’s appendix.
Appendix
Wildlife Monitoring
Wildlife monitoring uses sensors and data analysis to track animal species, abundance, behavior, and health for ecological research and conservation.
See all 68 papers