Fig 1.
An overview of the IMU-based dog gait analysis system.
(a) A photo of the IMU sensor and the container and a schematic of the sensor circuit. (b) A photo of the experimental setting. (c) A photo illustrating sensor position and axes orientation.
Fig 2.
Representative tracing from the sensor.
(a) Accelerometer readings. Signals are expressed as the g-force (9.81 m/s2). (b) Gyroscope readings. Signals are expressed as the angular velocity (deg/s). Shaded regions indicate swing phase of the limb.
Fig 3.
Method for detecting swing start and end.
(Top) Frame-by frame real-life images from video recording (images were mirrored so that the frame sequence matches the direction of motion and time). (Bottom right) Representative tracing of the z-axis angular velocity (Rz), x-axis acceleration (Ax) and the first derivative of Ax (Ax’). Circles indicate the Rz peak. The “cross” symbols indicate the start of swing and stance, respectively. (Bottom left) Charts depicting the histogram distribution of peaks in Rz and Ax’ for setting detection thresholds away from the signal baseline.
Fig 4.
A summary illustration of the step detection algorithm.
Fig 5.
Quantitative comparison of video results and sensor results for (a) swing start (toe-off event), (b) swing end (toe-touch event), (c) swing duration, (d) stance duration, and (e) stride duration.
Left and right panels depict the linear correlation and step-by-step error distribution, respectively, between the video and sensor results.
Fig 6.
The step phase duration as a function of Froude number and the distribution of duty factor in walk and trot gait types.
(a) The stance and swing durations at different Froude numbers. (b) The distribution of the duty factor for trot and walk gait types. The gait type (walk or trot) was identified in video analysis. The lines are best Gaussian fit: y = 0.23exp[-544.61(x-0.54)2], R2 = 0.98 for trot gait, and y = 0.12exp[-120.19(x-0.63)2], R2 = 0.94 for walk gait.