Validity of the Microsoft Kinect sensor for assessment of normal walking patterns in pigs

• Published in: Computers and electronics in agriculture Vol. 117; pp. 1 - 7
• Main Authors: Stavrakakis, Sophia, Li, Wei, Guy, Jonathan H., Morgan, Graham, Ushaw, Gary, Johnson, Garth R., Edwards, Sandra A.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2015
• Subjects: Algorithms; Automated detection; Kinect; Lameness; Markers; Motion capture; Necks; Pigs; Tracking; Trajectories; Walking; Walkways; Kinect; Pigs; Automated detection; Motion capture; Lameness
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2015.07.003

•Validity of Kinect for assessment of normal walking patterns in pigs was confirmed.•Kinect neck elevation trajectories agreed well with a gold standard.•The Kinect has potential to be developed into an automated lameness-detection tool. Lameness is a major problem affecting pigs and its detection is subjective and challenging on large farms. Previous research using advanced kinematic gait analysis (Vicon) has established that abnormality in the movement of the axial body during walking is associated with lameness in pigs. Vertical excursion of head and neck was most affected, and increased by +15–58mm in lame compared to normal pigs. However, simpler technology is required to automate lameness detection. In this experiment, walking trajectories of mid-line dorsal body regions of seven normal pigs varying in size were filmed repeatedly within day and between days on two or three occasions within one week. Trajectories were tracked simultaneously using both a 6-camera Vicon system, set up in an array flanking a walkway and detecting reflective markers, and a Microsoft Kinect motion sensor, mounted above the walkway. Four pigs wore a large (height 30mm) reflective marker in the mid-neck region, detectable by both Kinect and Vicon during two days. Two custom-written computer algorithms using the Kinect developer toolkit were produced to (1) follow the large neck marker and (2) enable marker-free tracking of other body regions. Reversed depth data from the Kinect and vertical position data from the Vicon were compared to assess agreement. There was a high positive correlation between the Kinect and Vicon trajectory means of the large neck marker (P<0.001; r=0.994). The Kinect neck marker trajectory mean was generally higher than the Vicon trajectory mean, therefore a positive difference of 4mm±4.2mm (LoA) was noted. There was no pig effect on trajectory differences, but a pig effect on trajectory mean which reflected the size of the pig (P<0.001). The mean±SD of continuous differences between corresponding Kinect and Vicon neck marker trajectories amounted to 5±1.5mm. The mean of vertical displacement amplitudes was 5±2.8mm, and hence the minimum difference of +15mm in lame animals should be detectable in more than 99% of cases. Trajectories of neck, back and pelvis generated by a marker-free Kinect application showed less similarity to corresponding Vicon trajectories. It was concluded that the Kinect device could distinguish sound from lame pigs by tracking neck region elevation during walking; however, markerfree tracking algorithms need refinement and further development to become sensitive and reliable.