The REALDISP (REAListic sensor DISPlacement) dataset has been originally collected to investigate the effects of sensor displacement in the activity recognition process in real-world settings. It builds on the concept of ideal-placement, self-placement and induced-displacement. The ideal and mutual-displacement conditions represent extreme displacement variants and thus could represent boundary conditions for recognition algorithms. In contrast, self-placement reflects a users perception of how sensors could be attached, e.g., in a sports or lifestyle application. The dataset includes a wide range of physical activities (warm up, cool down and fitness exercises), sensor modalities (acceleration, rate of turn, magnetic field and quaternions) and participants (17 subjects). Apart from investigating sensor displacement, the dataset lend itself for benchmarking activity recognition techniques in ideal conditions. ---------------------------------------------------------------------------------------------------------------------- Dataset summary: #Activities: 33 #Sensors: 9 #Subjects: 17 #Scenarios: 3 ---------------------------------------------------------------------------------------------------------------------- ACTIVITY SET: A1: Walking A2: Jogging A3: Running A4: Jump up A5: Jump front & back A6: Jump sideways A7: Jump leg/arms open/closed A8: Jump rope A9: Trunk twist (arms outstretched) A10: Trunk twist (elbows bent) A11: Waist bends forward A12: Waist rotation A13: Waist bends (reach foot with opposite hand) A14: Reach heels backwards A15: Lateral bend (10_ to the left + 10_ to the right) A16: Lateral bend with arm up (10_ to the left + 10_ to the right) A17: Repetitive forward stretching A18: Upper trunk and lower body opposite twist A19: Lateral elevation of arms A20: Frontal elevation of arms A21: Frontal hand claps A22: Frontal crossing of arms A23: Shoulders high-amplitude rotation A24: Shoulders low-amplitude rotation A25: Arms inner rotation A26: Knees (alternating) to the breast A27: Heels (alternating) to the backside A28: Knees bending (crouching) A29: Knees (alternating) bending forward A30: Rotation on the knees A31: Rowing A32: Elliptical bike A33: Cycling SENSOR SETUP: Each sensor provides 3D acceleration (accX,accY,accZ), 3D gyro (gyrX,gyrY,gyrZ), 3D magnetic field orientation (magX,magY,magZ) and 4D quaternions (Q1,Q2,Q3,Q4). The sensors are identified according to the body part on which is placed respectively: S1: left calf (LC) S2: left thigh (LT) S3: right calf (RC) S4: right thigh (RT) S5: back (BACK) S6: left lower arm (LLA) S7: left upper arm (LUA) S8: right lower arm (RLA) S9: right upper arm (RUA) SCENARIOS: The dataset contains information for three different scenarios depending on whether the sensors are positioned on predefined positions or placed by the users themselves. - Ideal-placement or the default scenario. The sensors are positioned by the instructor on predefined locations within each body part. The data stemming from this scenario could be considered as the “training set” for supervised activity recognition systems. - Self-placement. The user is asked to position a subset of the sensors themselves on the body parts specified by the instructor, but without providing any hint on how the sensors must be exactly placed. This scenario is devised to investigate some of the variability that may occur in the day-to-day usage of an activity recognition system, involving wearable or self-attached sensors. Normally, the self-placement will lead to on-body sensor setups that differ from the ideal-placement. Nevertheless, this difference may be minimal if the subject places the sensor close to the ideal position. - Induced-displacement. An intentional mispositioning of sensors using rotations and translations with respect to the ideal placement is introduced by the instructor. One of the key interests of including this last scenario is to investigate how the performance of a certain method degrades as the system drifts far from the ideal setup. A complete and illustrated description (including table of activities, sensor setup, etc.) of the dataset is provided in the documentation facilitated along with the dataset. Also, the papers presented in the section “Citation Requests” provide an insightful description of the dataset and the underlying theory.

No