Using visual lifelogs to automatically characterize everyday activities
Highlights
► Experimental evaluation of the efficacy of detecting activities from lifelog images. ► Positive performance results for automatic detection of basic human activities. ► Results of a field test on more than a dozen users. ► Automatically identifying human activities from a wearable camera has many uses.
Introduction
One of the most significant technological developments in our modern world is the development, and deployment, of various types of sensors throughout our society. As a result we can monitor, in an ambient way, many aspects of our lives and our environment. We are particularly interested in wearable sensors which include sensors to directly monitor human behavior as well as sensors in the mobile devices that we can carry around with us.
Lifelogging is the term used to describe the process of automatically, and ambiently, digitally recording our own day-to-day activities for our own personal purposes, using a variety of sensor types. This is opposed to having somebody else record what we are doing and using the logged data for some public or shared purpose. For example, an athlete recording his or her daily training and logging distance, time, etc. would count as a form of lifelog whereas a security firm monitoring a train station to detect anti-social behavior, would not.
One class of personal lifelogging called visual lifelogging is based on using wearable cameras, of which there are several examples now available including SenseCam [14], Looxcie®, GoPro®, Vicon Revue and the recently-announced Memento. These record either still images or video and are taken from a first-person view meaning they reflect the viewpoint that the wearer normally sees and usually they also record data from other wearable sensors. Applications for visual lifelogging are manyfold. Sometimes they are job- or task-related, sometimes we lifelog for leisure, and more recently we’re seeing lifelogging used for health applications as well as applications for real time lifelogging [34].
In this paper we use visual lifelogging in the task of characterizing the activities of the user of a wearable camera, SenseCam. The major contribution of this paper is the proposal of HMM-based models to characterize everyday activities by merging time-varying dynamics of high-level features (concepts). Utilizing concept-concept relationships to map concept vectors to a more compact space by LSA and the extensive experiments on various concept detection performances generated by Monte Carlo simulations are another contribution of this paper, which have not been reported before in semantic analysis of visual lifelogging, to the best of our knowledge. We describe related work in lifelogging followed by an overview of the approaches to managing lifelog data. In Section 3 we introduce the background to our experiments including how we define a vocabulary of 85 base semantic concepts. Section 4 presents a Hidden Markov Model approach to deal with the activities to be detected. We then present our experiments using both ‘clean’ or perfect annotation as well as automatic detection of base semantic concepts, followed by an analysis of two different sampling methods. We conclude the paper with a re-cap on our contribution and directions for future work.
Section snippets
Related work
Lifelogging is a very broad topic both in the technologies that can be used, and the applications for lifelogged data. For the most part, lifelogging applications are based around health and wellness, though we have seen applications as diverse as theater and dance [48]. We describe related work in visual lifelogging where we broadly divide this into applications for memory recall and applications for lifestyle analysis, though new application areas are emerging.
The seminal work in visual
Managing lifelog data
The application of lifelogging, especially visual lifelogging, to analysis of the activities of the wearer creates challenging problems for retrieval due to the large volume of lifelog data and the fact that much of that data is repetitive with repeated images of the same or nearly the same thing. Recording every activity of a wearer’s life will generate a large amount of data for a typical day, not to say for a longer term, for example, a month or even a year. Detecting events or activities
Experimental setup and variables
In this paper, the methodology we proposed for the investigation of everyday activity characterization can be demonstrated by the algorithm pipeline shown in Fig. 2. The algorithm consists of four main components which are concept identification from raw SenseCam images, vocabulary construction for visual semantics, the modeling of time-varying patterns by HMM and activity classification through trained HMM models. The vocabulary construction module can further be boiled down to LSA (Latent
Evaluation data set
In the experiment on evaluating activity classification, we carried out an assessment of our algorithm on data sets using both clean (correct) concept annotation and on concept annotation with errors. The data sets we used are event samples of the 23 activity types from Section 3.3 collected by 4 people with different demographics (older people vs. younger researchers), one older participant who is less functional in terms of capacity for household and community activities from an occupational
Varying the sampling method
As described in Section 5.1, each event sample is divided into two halves, of which the first half is used as training data and the other is used as testing data. To evaluate the effect of this sampling method for training and test data, we also carried out the same experiment on another sampling method, odd-and-even sampling, to distinguish from half-and-half sampling. That is, in each event sample, we used the odd numbered images as training data while the images with even number are used as
Conclusions
In this paper we have described a novel application of visual lifelogging where a subject wears a camera that records images of their day-to-day activities, ambiently. Our particular interest is in characterizing the activities and everyday behavior of the wearer which is distinct from other applications of visual lifelogging like remembrance or re-finding previous events from the past. The novelty of our contribution lies in the fact we have used visual images as the raw source of user
Authors’ contributions
Peng Wang carried out the experiments under the supervision of Alan Smeaton. Both authors wrote the paper with equal contribution and both authors approve this submission.
Conflict of interests
The authors have no conflicts of interest in undertaking or reporting this research.
Acknowledgements
This work was supported by Science Foundation Ireland under Grant No. 07/CE/I1147 and by the China Scholarship Council, neither of which had any involvement in the work, in writing this article or in deciding on its submission to this journal.
References (51)
- et al.
Beyond shot retrieval: searching for broadcast news items using language models of concepts
- et al.
Concept detectors: how good is good enough?
- et al.
Simulating the future of concept-based video retrieval under improved detector performance
Multimedia Tools and Applications
(2011) - et al.
The use of a wearable camera, sensecam, as a pictorial diary to improve autobiographical memory in a patient with limbic encephalitis: a preliminary report
Neuropsychological Rehabilitation
(2007) - et al.
Sensecam improves memory for recent events and quality of life in a patient with memory retrieval difficulties
Memory
(2011) - et al.
Investigation into the occupational lives of healthy older people through their use of time
Australian Occupational Therapy Journal
(2010) - et al.
Combining inertial and visual sensing for human action recognition in tennis
- et al.
Indexing by latent semantic analysis
Journal of the American Society for Information Science
(1990) - et al.
Passively recognising human activities through lifelogging
Computers in Human Behavior
(2011) - et al.
Visual event detection using multi-dimensional concept dynamics