תפריט נגישות
ניגודיות עדינהניגודיות גבוהה
מונוכרום
הדגשת קישורים
חסימת אנימציה
פונט קריא
סגור
איפוס הגדרות נגישות
להורדת מודול נגישות חינם תפריט נגישותניגודיות עדינהניגודיות גבוההמונוכרוםהדגשת קישוריםחסימת אנימציהפונט קריאסגוראיפוס הגדרות נגישותלהורדת מודול נגישות חינםhe objective of this study was to evaluate the system performance of a 3D vision system for automatic locomotion monitoring implemented in a commercial dairy farm. Data were gathered during 633 milking sessions on a Belgian commercial dairy farm. After milking, the cows walked in a single-lane alley where the video recording system with a 3D depth camera was installed. The entire monitoring process including video recording, video pre-processing by filtering, cow identification and video analysis was automated. Image processing extracted six feature variables from the recorded videos. Per milking session, 224 ± 10 cows (100%) were identified on average by a radio-frequency identification (RFID) antenna, and 197 ± 16 videos were recorded (88.1 ± 6.6%) by the camera. The cow identification number was merged automatically to a recorded video in 178 ± 14 videos (79.4 ± 5.5%). After video pre-processing and analysis, 110 ± 24 recorded cow-videos (49.3 ± 10.8%) per session resulted in an automatic locomotion score. Daily and cow-individual variations on the merging and analysis rate were due to cow traffic. The minimal cow traffic interval required between consecutive cows was 15 s for optimal merging. System performance was affected by lactation stage, parity of the cows and recording duration. The feature variables curvature angle of back around hip joints (Area Under the Receiver Operating Characteristics Curve (AUC) = 0.719) and back posture measurement (AUC = 0.702) could be considered as fair lameness classifiers. Cow traffic affected the success rate of the video processing. Therefore, automatic monitoring systems need to be adapted to the farm layout.
he objective of this study was to evaluate the system performance of a 3D vision system for automatic locomotion monitoring implemented in a commercial dairy farm. Data were gathered during 633 milking sessions on a Belgian commercial dairy farm. After milking, the cows walked in a single-lane alley where the video recording system with a 3D depth camera was installed. The entire monitoring process including video recording, video pre-processing by filtering, cow identification and video analysis was automated. Image processing extracted six feature variables from the recorded videos. Per milking session, 224 ± 10 cows (100%) were identified on average by a radio-frequency identification (RFID) antenna, and 197 ± 16 videos were recorded (88.1 ± 6.6%) by the camera. The cow identification number was merged automatically to a recorded video in 178 ± 14 videos (79.4 ± 5.5%). After video pre-processing and analysis, 110 ± 24 recorded cow-videos (49.3 ± 10.8%) per session resulted in an automatic locomotion score. Daily and cow-individual variations on the merging and analysis rate were due to cow traffic. The minimal cow traffic interval required between consecutive cows was 15 s for optimal merging. System performance was affected by lactation stage, parity of the cows and recording duration. The feature variables curvature angle of back around hip joints (Area Under the Receiver Operating Characteristics Curve (AUC) = 0.719) and back posture measurement (AUC = 0.702) could be considered as fair lameness classifiers. Cow traffic affected the success rate of the video processing. Therefore, automatic monitoring systems need to be adapted to the farm layout.
