Adapting the re-ID challenge for static sensors

\au Avirath Sundaresan ​ 1 , † {}^{\text{\hskip 2.0pt}1,{\dagger}} \au Jason Parham ​ 2 , † {}^{\text{\hskip 2.0pt}2,{\dagger}} \au Jonathan Crall ​ 2 {}^{\text{\hskip 3.0pt}2} \au Rosemary Warungu ​ 5 {}^{\text{\hskip 2.0pt}5} \au Timothy Muthami ​ 5 {}^{\text{\hskip 3.0pt}5} \au Jackson Miliko ​ 5 · 2024

Read Adapting the re-ID challenge for static sensors by \au Avirath Sundaresan ​ 1 , † {}^{\text{\hskip 2.0pt}1,{\dagger}} \au Jason Parham ​ 2 , † {}^{\text{\hskip 2.0pt}2,{\dagger}} \au Jonathan Crall ​ 2 {}^{\text{\hskip 3.0pt}2} \au Rosemary Warungu ​ 5 {}^{\text{\hskip 2.0pt}5} \au Timothy Muthami ​ 5 {}^{\text{\hskip 3.0pt}5} \au Jackson Miliko ​ 5 (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-00290.txt.

Introduction

Avirath Sundaresan and a large team of co-authors adapt animal re-identification to camera traps that never move. The work grows out of counts of Grevy's zebras, a highly endangered species tallied in citizen science rallies in 2016 and 2018. Fixed camera networks could watch the population all year, but most wild photos are too poor for matching an animal's stripe pattern. Their 2024 paper reframes the task for static sensors, where the same spots are seen again and again.

Knowing the place and angle of each camera adds clues that rally photos lack. The team studies how to use those clues and how to handle the many unusable frames. Long-term, cheap monitoring would help protect a species counted in the low thousands. The method may extend to other patterned animals.

Catalog record

CREATOR\au Avirath Sundaresan ​ 1 , † {}^{\text{\hskip 2.0pt}1,{\dagger}} \au Jason Parham ​ 2 , † {}^{\text{\hskip 2.0pt}2,{\dagger}} \au Jonathan Crall ​ 2 {}^{\text{\hskip 3.0pt}2} \au Rosemary Warungu ​ 5 {}^{\text{\hskip 2.0pt}5} \au Timothy Muthami ​ 5 {}^{\text{\hskip 3.0pt}5} \au Jackson Miliko ​ 5
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-00290

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Adapting the re-ID challenge for static sensors \au Avirath Sundaresan ​ 1 , † {}^{\text{\hskip 2.0pt}1,{\dagger}} \au Jason Parham ​ 2 , † {}^{\text{\hskip 2.0pt}2,{\dagger}} \au Jonathan Crall ​ 2 {}^{\text{\hskip 3.0pt}2} \au Rosemary Warungu ​ 5 {}^{\text{\hskip 2.0pt}5} \au Timothy Muthami ​ 5 {}^{\text{\hskip 3.0pt}5} \au Jackson Miliko ​ 5 {}^{\text{\hskip 1.0pt}5} \au Margaret Mwangi ​ 5 {}^{\text{\hskip 3.0pt}5} \au Jason Holmberg ​ 4 {}^{\text{\hskip 2.0pt}4} \au Tanya Berger-Wolf ​ 6 , 4 {}^{\text{\hskip 3.0pt}6,4} \au Daniel Rubenstein ​ 7 {}^{\text{\hskip 2.0pt}7} \au Charles Stewart ​ 3 , 4 {}^{\text{\hskip 2.0pt}3,4} \au Sara Beery ​ 8 , ∗ {}^{\text{\hskip 2.0pt}8,*} Abstract The Grévy’s zebra, an endangered species native to Kenya and southern Ethiopia, has been the target of sustained conservation efforts in recent years. Accurately monitoring Grévy’s zebra populations is essential for ecologists to evaluate ongoing conservation initiatives. Recently, in both 2016 and 2018, a full census of the Grévy’s zebra population has been enabled by the Great Grévy’s Rally (GGR), a citizen science event that combines teams of volunteers to capture data with computer vision algorithms that help experts estimate the number of individuals in the population. A complementary, scalable, cost-effective, and long-term Grévy’s population monitoring approach involves deploying a network of camera traps, which we have done at the Mpala Research Centre in Laikipia County, Kenya. In both scenarios, a substantial majority of the images of zebras are not usable for individual identification, due to "in-the-wild" imaging conditions — occlusions from vegetation or other animals, oblique views, low image quality, and animals that appear in the far background and are thus too small to identify. Camera trap images, without an intelligent human photographer to select the framing and focus on the animals of interest, are of even poorer quality, with high rates of occlusion and high spatio-temporal similarity within image bursts. We employ an image filtering pipeline incorporating animal detection, species identification, viewpoint estimation, quality evaluation, and temporal subsampling to compensate for these factors and obtain individual crops from camera trap and GGR images of suitable quality for re-ID. We then employ the Local Clusterings and their Alternatives (LCA) algorithm, a hybrid computer vision & graph clustering method for animal re-ID, on the resulting high-quality crops. Our method processed images taken during GGR-16 and GGR-18 in Meru County, Kenya, into 4,142 highly-comparable annotations, requiring only 120 contrastive same-vs-different-individual decisions from a human reviewer to produce a population estimate of 349 individuals (within 4.6 % \% of the ground-truth count in Meru County). Our method also efficiently processed 8.9M unlabeled camera trap images from 70 camera traps at Mpala over two years into 685 encounters of 173 unique individuals, requiring only 331 contrastive decisions from a human reviewer. \supertitle 1 Introduction The population of Grévy’s zebras experienced a dramatic decline beginning in the 1970s largely due to hunting and competition for food and water resources with local pastoral communities. Estimates have placed the number of Grévy’s zebras remaining in the wild at under 2,000, with the vast majority in the Samburu region of central Kenya. Due to extensive conservation efforts by the Kenyan and Ethiopian governments as well as environmental NGOs, the population of Grévy’s zebras have stabilized in recent years Rubenstein et al. [2016] . Accurately censusing Grévy’s zebra populations is critical for ecologists to evaluate these existing conservation efforts. Developing and maintaining a census of known individuals requires effective animal re-identification methods to ensure that only unique individuals are included in population counts Tuia et al. [2022] . A popular method for population size estimation is “capture-mark-recapture" Pradel [1996] . By this method, a set of animals in the target population is first captured and marked, then a second set of animals is independently recaptured, and finally a population estimate formed from the number of animals captured twice. This method, however, proves difficult to scale to large populations and territories, and may lead to inaccurate population estimates when animals are not confined to the study area and are capable of evading tagging Lee et al. [2014] ; Tilling and Sterne [1999] . Further, the tagging methods Figure 1: Map of several encounters of a single individual within Mpala Research Centre identified from camera trap data by our method and requiring minimal human labeling. used in population estimation studies (such as ear tags and radio collars Alexander and Appel [1994] ; Thouless [1995] ; Collins et al. [2014] ) can be excessively expensive and time-consuming to implement in the field, and overly harmful to the animal Witmer [2005] . A modern alternative to manual mark-recapture studies that does not have these limitations is exemplified in the Great Grévy’s Rally (GGR) events of 2016 and 2018 where volunteers spread over the range of Grévy’s zebra to photograph as many animals as possible over two consecutive days. Relying on the distinctive appearance of an animal itself as a means for identification, a combination of algorithmic and human curation efforts produced population estimates that have been accepted as definitive by the Kenyan Wildlife Service and the IUCN Rubenstein et al. [2018a] . Despite the success of the GGR, the curation effort of ˜50,000 images per event is formidable, since the dataset must be narrowed to the recognizable subset of images of Grévy’s and a streamlined curation procedure must be developed, all-the-while preserving the accuracy of the resulting census. Parham [2021] demonstrates how to ensure accuracy while reducing human effort. Figure 2: (Left) Location of our camera traps across the Mpala Research Centre. These cameras are placed according to four distinct strategies: random grid, randomly along roads, at known Grévy’s territories along roads, and at magnets such as salt licks and watering holes. (Right) Map of all Grévy’s encounters across Mpala. Complementary to the focused burst of human effort for the GGR photographic events is the possibility of using a fleet of camera traps to collect the data needed for population monitoring. For animals that allow for sight-based ID, images taken by camera traps are a potentially cost-effective and non-invasive method to re-identify individuals and obtain robust population estimates. Unlike alternative methods such as tagging or even the field photography of the GGR events, camera traps do not require the physical presence of field researchers for animal re-identification; this ensures that the natural behaviors of animals are not disrupted and saves considerable time and resources Delisle et al. [2021] ; Schneider et al. [2019] . However, analysis of camera trap images by human observation alone is not practical. In recent years, computer vision algorithms have been shown to present a highly accurate and standardized method for camera trap image analysis. Computer vision techniques have seen significant success in automated species detection & identification from camera trap images, especially with advances in deep learning Norouzzadeh et al. [2019] ; Beery et al. [2019a] ; Beery et al. [2019b] ; Beery et al. [2019c] ; Tabak et al. [2020] . Accurate species ID from camera trap images has laid the foundation for the next step in an end-to-end photographic censusing pipeline from camera trap images: automated individual re-ID (for reviews, see: Schneider et al. [2019] ; Ravoor and T. S. B. [2020] ; Vidal et al. [2021] ). However, the challenging nature of

Read the complete text