Surveillance system and method for predicting patient falls using motion feature patterns
A method and system for detecting a fall risk condition, the system comprising a surveillance camera configured to generate a plurality of frames showing an area in which a patient at risk of falling is being monitored, and a computer system comprising memory and logic circuitry configured to store motion feature patterns that are extracted from video recordings, the motion feature patterns are representative of motion associated with real alarm cases and false-alarm cases of fall events, receive a fall alert from a classifier, determine motion features of one or more frames from the plurality of frames that correspond to the fall alert; compare the motion features of the one or more frames with the motion feature patterns, and determine whether to confirm the fall alert based on the comparison.
1. A surveillance system for detecting a fall risk condition, the system comprising:
a computer system comprising memory and logic circuitry configured to:
retrieve, from a storage device, video data that is associated with triggering false alarm cases of patient falls;
determine motion feature patterns that are associated with risk or non-risk events from the video data;
classify, using a classifier that is trained with a set of features associated with triggering alarms for patient falls, one or more of a plurality of frames generated from a surveillance camera as causing an alarm;
verify the classification by extracting motion feature patterns of the one or more of the plurality of frames and comparing the extracted motion feature patterns with the determined motion feature patterns; and
generate feedback to the classifier with training data based on the verification of the alarm.
2. The surveillance system of claim 1 wherein the video data includes a log of events.
3. The surveillance system of claim 1 further comprising the computer system determining real-alarm cases of patient falls from the video data based on triggering of consecutive alarms.
4. The surveillance system of claim 1 further comprising the computer system receiving graphically defined areas to monitor for patient falls through a user interface.
5. The surveillance system of claim 4 wherein the determined motion feature patterns correspond to motion of a virtual bed zone that is generated based on the graphically defined areas.
6. The surveillance system of claim 1 wherein the training data includes the determined motion feature patterns.
7. The surveillance system of claim 1 wherein the motion feature patterns include at least one of a centroid location, centroid area, connected components ratio, bed motion percentage, and unconnected motion.
8. A method for detecting a fall risk condition, the method comprising:
retrieving, by a computer system from a storage device, video data that is associated with triggering false alarm cases of patient falls;
determining, by the computer system, motion feature patterns that are associated with risk or non-risk events from the video data;
classifying, by the computer system using a classifier that is trained with a set of features associated with triggering alarms for patient falls, one or more of a plurality of frames generated from a surveillance camera as causing an alarm;
verifying, by the computer system, the classification by extracting motion feature patterns of the one or more of the plurality of frames and comparing the extracted motion feature patterns with the determined motion feature patterns; and
generating, by the computer system, feedback to the classifier with training data based on the verification of the alarm.
9. The method of claim 8 wherein the video data includes a log of events.
10. The method of claim 8 further comprising determining real-alarm cases of patient falls from the video data based on triggering of consecutive alarms.
11. The method of claim 8 further comprising receiving graphically defined areas to monitor for patient falls through a user interface.
12. The method of claim 11 wherein the determined motion feature patterns correspond to motion of a virtual bed zone that is generated based on the graphically defined areas.
13. The method of claim 8 wherein the training data includes the determined motion feature patterns.
14. The method of claim 8 wherein the motion feature patterns include at least one of a centroid location, centroid area, connected components ratio, bed motion percentage, and unconnected motion.
15. Non-transitory computer-readable media comprising program code that when executed by a programmable processor causes execution of a method for detecting a fall risk condition, the computer-readable media comprising:
computer program code for retrieving from a storage device, video data that is associated with triggering false alarm cases of patient falls;
computer program code for determining motion feature patterns that are associated with risk or non-risk events from the video data;
computer program code for classifying, using a classifier that is trained with a set of features associated with triggering alarms for patient falls, one or more of a plurality of frames generated from a surveillance camera as causing an alarm;
computer program code for verifying the classification by extracting motion feature patterns of the one or more of the plurality of frames and comparing the extracted motion feature patterns with the determined motion feature patterns; and
computer program code for generating feedback to the classifier with training data based on the verification of the alarm.
16. The non-transitory computer-readable media of claim 15 further comprising computer program code for determining real-alarm cases of patient falls from the video data based on triggering of consecutive alarms.
17. The non-transitory computer-readable media of claim 15 further comprising computer program code for receiving graphically defined areas to monitor for patient falls through a user interface.
18. The non-transitory computer-readable media of claim 17 wherein the determined motion feature patterns correspond to motion of a virtual bed zone that is generated based on the graphically defined areas.
19. The non-transitory computer-readable media of claim 15 wherein the training data includes the determined motion feature patterns.
20. The non-transitory computer-readable media of claim 15 wherein the motion feature patterns include at least one of a centroid location, centroid area, connected components ratio, bed motion percentage, and unconnected motion.