DeepWILD: Wildlife Identification,Localisation and estimation on camera trap videos using Deep learning

Videos and images from camera traps are more and more used by ecologists to estimate the population of species on a territory. It is a laborious work since experts have to analyse massive data sets manually. This takes also a lot of time to filter these videos when many of them do not contain animals or are with human presence. Fortunately, deep learning algorithms for object detection can help ecologists to identify multiple relevant species on their data and to estimate their population. In this study, we propose to go even further by using object detection model to detect, classify and count species on camera traps videos. To this end, we developed a 3-step process: (i) At the first stage, after splitting videos into images, we annotate images by associating bounding boxes to each label thanks to MegaDetector algorithm; (ii) then, we extend MegaDetector based on Faster R-CNN architecture with backbone Inception-ResNet-v2 in order to not only detect the 13 relevant classes but also to classify them; (iii) finally, we design a method to count individuals based on the maximum number of bounding boxes detected. This final stage of counting is evaluated in two different contexts: first including only detection results (i.e. comparing our predictions against the right number of individuals, no matter their true class), then an evolved version including both detection and classification results (i.e. comparing our predictions against the right number in the right class). The results obtained during the evaluation of our model on the test data set are: (i) 73,92% mAP for classification, (ii) 96,88% mAP for detection with a ratio Intersection-Over-Union (IoU) of 0.5 (overlapping ratio between groundtruth bounding box and the detected one), and (iii) 89,24% mAP for detection at IoU = 0.75. Highly represented classes, like humans, have highest values of mAP around 81% whereas less represented classes in the train data set, such as dogs, have lowest values of mAP around 66%. Regarding the proposed counting method, we predicted a count either exact or $\pm$ 1 unit for 87% with detection results and for 48% with detection and classification results of our test data set. Our model is also able to detect empty videos. To the best of our knowledge, this is the first study in France about the use of object detection model on a French national park to locate, identify and estimate the population of species from camera trap videos.     

Accuracy of identifications of mammal species from camera trap images: A northern Australian case study

Camera traps are a powerful and increasingly popular tool for mammal research, but like all survey methods, they have limitations. Identifying animal species from images is a critical component of camera trap studies, yet while researchers recognize constraints with experimental design or camera technology, image misidentification is still not well understood. We evaluated the effects of a species’ attributes (body mass and distinctiveness) and individual observer variables (experience and confidence) on the accuracy of mammal identifications from camera trap images. We conducted an Internet‐based survey containing 20 questions about observer experience and 60 camera trap images to identify. Images were sourced from surveys in northern Australia and included 25 species, ranging in body mass from the delicate mouse (Pseudomys delicatulus, 10 g) to the agile wallaby (Macropus agilis, >10 kg). There was a weak relationship between the accuracy of mammal identifications and observer experience. However, accuracy was highest (100%) for distinctive species (e.g. Short‐beaked echidna [Tachyglossus aculeatus]) and lowest (36%) for superficially non‐distinctive mammals (e.g. rodents like the Pale field‐rat [Rattus tunneyi]). There was a positive relationship between the accuracy of identifications and body mass. Participant confidence was highest for large and distinctive mammals, but was not related to participant experience level. Identifications made with greater confidence were more likely to be accurate. Unreliability in identifications of mammal species is a significant limitation to camera trap studies, particularly where small mammals are the focus, or where similar‐looking species co‐occur. Integration of camera traps with conventional survey techniques (e.g. live‐trapping), use of a reference library or computer‐automated programs are likely to aid positive identifications, while employing a confidence rating system and/or multiple observers may lead to a collection of more robust data. Although our study focussed on Australian species, our findings apply to camera trap studies globally.