Dealing with sensor displacement in motion-based onbody activity recognition systems

Abstract

We present a set of heuristics that significantly increase the robustness of motion sensor-based activity recognition with respect to sensor displacement. In this paper placement refers to the position within a single body part (e.g, lower arm). We show how, within certain limits and with modest quality degradation, motion sensorbased activity recognition can be implemented in a displacement tolerant way. We first describe the physical principles that lead to our heuristic. We then evaluate them first on a set of synthetic lower arm motions which are well suited to illustrate the strengths and limits of our approach, then on an extended modes of locomotion problem (sensors on the upper leg) and finally on a set of exercises performed on various gym machines (sensors placed on the lower arm). In this example our heuristic raises the displaced recognition rate from 24% for a displaced accelerometer, which had 96% recognition when not displaced, to 82%.

Collecting complex activity datasets in highly rich networked sensor environments

“…Thus, classification methods must be robust to possible signal variations. We showed how activity recognition can be made resilient to small changes in on-body sensor placement using unsupervised techniques [18], principles of body mechanics [19], or evolutionary techniques [20]. We showed that sensors can autonomously recognize their on-body position [21] and their symbolic location in the environment [22].…”

Benchmarking classification techniques using the Opportunity human activity dataset

“…due to changes in the user's behavior or as a result of noise). Therefore, several challenges arise at the different processing stages from the feature selection and classification [6], [7], to sensor and decision fusion [8], as well as fault-tolerance [9], [10], [11]. Moreover, real-life deployments are required to detect when no relevant action is performed (i.e.…”

Robust activity recognition combining anomaly detection and classifier retraining

“…For example, in the case of sensor displacement or rotation, the first approach, followed by [6], has the disadvantage of losing the information on the orientation of the sensors, which can be necessary for recognizing fine-grained activities or gestures. Extracting robust features with respect to certain type of anomaly (sensor displacement) is studied in [7], with extra burden on feature extraction phase and may not be generalized to other type of anomalies.…”