Accelerometers, Ground Truthing, and Supervised Learning
Accelerometers are sensitive to movement and the lack of it. They are not sentient and must recognise animal behaviour based on a human observer’s cognition. Therefore, remote recognition of behaviour using accelerometers requires ground truth data which is based on human observation or knowledge. The need for validated behavioural information and for automating the analysis of the vast amounts of data collected today, have resulted in many studies opting for supervised machine learning approaches.
In such approaches, the process of ground truthing involves time-synchronising acceleration signals with simultaneously recorded video, having an animal behaviour expert create an ethogram, and then annotate the video according to this ethogram. This links the recorded acceleration signal to the stream of observed animal behaviours that produced it. After this, acceleration signals are chopped up into finite sections of pre-set size (e.g. two seconds), called windows. From acceleration data within windows, quantities called ‘features’ are engineered with the aim of summarising characteristics of the acceleration signal. Typically, ~15-20 features are computed. Good features will have similar values for the same behaviour, and different values for different behaviours.
It’s more important than ever for us to have accurate information to help marine conservation efforts. Jordan Goetze and his colleagues have provided the first comprehensive guide for researchers using diver operated stereo-video methods (or stereo-DOVs) to survey fish assemblages and their associated habitat.
But what is Stereo DOV? What makes it a better method than the traditional UVC (Underwater Visual Census) method? And when should you use it? Find out in this video:
Quantifying animal movement is central to research spanning a variety of topics. It’s an important area of study for behavioural ecologists, evolutionary biologists, ecotoxicologists and many more. There are a lot of ways to track animals, but they’re often difficult, especially for people who don’t have a strong background in programming.
Vivek Hari Sridhar, Dominique G. Roche and Simon Gingins have developed a new, simple software to help with this though: Tracktor. This package provides researchers with a free, efficient, markerless video-based tracking solution to analyse animal movement of single individuals and groups.
Vivek and Simon explain the features and strengths of Tracktor in this new video:
Anyone who studies social animals in the wild (or human groups, for that matter), will soon find that some individuals threaten or attack others frequently, while others try to get out of the way or signal their submission in response to aggression. Observers tally the outcome of such aggressive interactions between any given two individuals (or ‘dyads’) and try to deduce the rank hierarchy from such winner-loser matrices. One drawback of this approach is that all temporal information is lost.
Imagine Royal, a baboon, dominating over Power, another baboon, 20 times, and Power dominating over Royal 20 times as well. If we just look at these data, we might think that they have the same fighting ability and similar ranks. But, if we know that Royal beat Power the first 20 of the interactions, then Power beat Royal in all further interactions, we’d come to a totally different conclusion. We’d infer that Power had toppled Royal and a rank change had taken place.
How do Rank Hierarchies Change Over Time?
One prominent method that takes the temporal dynamics of winner-loser interactions into account was originally developed to calculate the relative skill level of chess players. This method was introduced by Arpad Elo and is hence known as Elo-Rating. Elo-Rating has also been applied to rate the relative skills in a variety of competitive fields, including Major League Baseball, video games, and Scrabble. Continue reading →
Motion vision is an important source of information for many animals. It facilitates an animal’s movement through an environment, as well as being essential for locating prey and detecting predators. However, information on the conditions for motion vision in natural environments is limited.
To address this, Bian et al. have developed an innovative approach that combines novel field techniques with tools from 3D animation to determine how habitat structure, weather and motion vision influence animal behaviour. Their project focuses on Australia’s charismatic dragon lizards, and will place the animals’ motion displays in a visual-ecological context. The application of this approach goes well beyond this topic and the authors suggest the motion graphic technologies is a valuable tool for investigating the visual ecology of animals in a range of environments and at different spatial and temporal scales.
Many animals rely on movement to find prey and avoid predators. Movement is also an essential component of the territorial displays of lizards, comprising tail, limb, head and whole-body movements.
For the first time, digital animation has been used as a research tool to examine how the effectiveness of a lizard’s territorial display varies across ecological environments and conditions. The new research was published today in the journal Methods in Ecology and Evolution.
Around the world there are concerns over the impacts of land use change and the developments (such as wind farms). These concerns have led to the implementation of tracking studies to better understand movement patterns of animals. Such studies have provided a wealth of high-resolution data and opportunities to explore sophisticated statistical methods for analysis of animal behaviour.
We use accelerometer data from aerial (Verreaux’s eagle in South Africa) and marine (blacktip reef shark in Hawai’i) systems to demonstrate the use of hidden Markov models (HMMs) in providing quantitative measures of behaviour. HMMs work really well for analysing animal accelerometer data because they account for serial autocorrelation in data. They allow for inferences to be made about relative activity and behaviour when animals cannot be directly observed too, which is very important.
In addition to this, HMMs provide data-driven estimates of the underlying distributions of the acceleration metrics – and the probability of switching between states – possibly as a function of covariates. The framework that we provide in ‘Analysis of animal accelerometer data using hidden Markov models‘ can be applied to a wide range of activity data. It opens up exciting opportunities for understanding drivers of individual animal behaviour.
Verreaux’s eagles lay one to two eggs but only raise one chick. These are normally hatch during mid-winter and stay on the nest for around 90 days. The parents provision the chick during this time, delivering prey to the nest. Rock hyrax is a frequent item on the menu, as seen here. Having concurred data from accelerometers and nest cameras for the same bird could allow for activity level to be linked to successful prey acquisition or prey type. (Photo taken by a nest camera installed as part of the Black Eagle Project, Megan Murgatroyd)
“I am broadly interested in developing and applying statistical tools to infer behavioural and population processes from empirical data. My work tends to focus on marine and polar mammals, but the methods I develop are often applicable to a wide range of species and ecosystems. My recent work has centred on modelling animal behaviour using movement data and I generally analyse data with spatial and/or temporal structure.”