Computer-vision based workplace safety
View Patent ↗An example provides a method, including: obtaining, using one or more sensor systems, sensor data for one or more persons in an environment comprising machinery; analyzing, using a processor, the sensor data using a trained model to identify the one or more persons and an associated time series of actions; determining, using the associated time series of actions, if the one or more persons are engaging in behavior indicative of an unsafe practice; and thereafter presenting safety information to a visual display system, connected software system or other decision support tool.
1. A method, comprising:
obtaining, using one or more sensor systems, sensor data for one or more persons in an environment comprising machinery;
analyzing, using a processor, the sensor data comprising a plurality of images, each image having a temporal characteristic, the analyzing being based on a trained model comprising a temporal classifier configurable to:
identify, based on one or more spatial characteristics, two or more images from the plurality of images, each of the two or more images showing the one or more persons engaging in at least one action indicative of engagement with the machinery;
determine, based on temporal characteristics of the two or more images, that the at least one action is associated with a composite event; and
determine, based on the one or more spatial characteristics, that the one or more persons are performing the composite event in behavior indicative of an unsafe practice; and
thereafter presenting safety information to a visual display system, connected software system or other decision support tool;
wherein to determine that the one or more persons are performing the composite event in the behavior indicative of the unsafe practice is based on a Safety Activity Threshold (SAT);
wherein the sensor data further comprises machine state data, and wherein the SAT is contextually adjusted based on the machine state data; and
wherein the processor contextually adjusts the SAT to indicate an increased probability that the one or more persons are performing the composite event in the behavior indicative of the unsafe practice responsive to the machine state data indicating that the machinery is active.
2. The method of claim 1 , wherein the sensor data is a combination of one two or more of closed-circuit television, RGB image data, infrared data, mm-wave data, THz spectroscopy, audio signals, radar signals and LIDAR signals.
3. The method of claim 1 , wherein the sensor data is obtained from a single camera system with multiple modes of operation as a multispectral imager.
4. The method of claim 1 , wherein sensor data is collected from a network of distributed sensors throughout the environment, including one or more of mobile phones, desktop computers or single-purpose embedded devices.
5. The method of claim 1 , wherein the safety information is presented on the visual display system in real time.
6. The method of claim 1 , wherein the behavior indicative of an unsafe practice is stored in a database and updated continuously over time.
7. The method of claim 1 , comprising receiving operator feedback relating to the presented safety information.
8. A system, comprising:
one or more sensors;
one or more processors; and
a non-transitory storage device operatively coupled to the one or more sensors and the one or more processors, the non-transitory storage device comprising code executable by the one or more processors and configured to:
obtain sensor data for one or more persons in an environment comprising machinery;
analyze the sensor data comprising a plurality of images, each image having a temporal characteristic, based on a trained model comprising a temporal classifier configurable to:
identify, based on one or more spatial characteristics, two or more images from the plurality of images, each of the two or more images showing the one or more persons engaging in at least one action indicative of engagement with the machinery;
determine, based on temporal characteristics of the two or more images, that the at least one action is associated with a composite event; and
determine, based on the one or more spatial characteristics, that the one or more persons are performing the composite event in behavior indicative of an unsafe practice; and
thereafter provide safety information to a visual display system, connected software system or other decision support tool;
wherein to determine that the one or more persons are performing the composite event in the behavior indicative of the unsafe practice is based on a Safety Activity Threshold (SAT);
wherein the sensor data further comprises machine state data, and wherein the SAT is contextually adjusted based on the machine state data; and
wherein the one or more processors contextually adjusts the SAT to indicate an increased probability that the one or more persons are performing the composite event in the behavior indicative of the unsafe practice responsive to the machine state data indicating that the machinery is active.
9. The system of claim 8 , wherein the sensor data is a combination of one or more of closed-circuit television, RGB image data, infrared data, mm-wave data, THz spectroscopy, audio signals, radar signals and LIDAR signals.
10. The system of claim 8 , wherein the sensor data is obtained from a single camera system with multiple modes of operation as a multispectral imager.
11. The system of claim 8 , wherein sensor data is collected from a network of distributed sensors throughout the environment, including one or more of mobile phones, desktop computers or single-purpose embedded devices.
12. The system of claim 8 , wherein the safety information is presented on the visual display system in real time.
13. The system of claim 8 , wherein a determination that the one or more persons are performing the composite event in the behavior indicative of an unsafe practice is stored in a database and updated continuously over time.
14. A computer program product, comprising:
a non-transitory storage device comprising code executable by a processor and configured to:
obtain sensor data for one or more persons in an environment comprising machinery;
analyze a time series of images of the sensor data comprising a plurality of images, each image having a temporal characteristic, based on a trained model comprising a temporal classifier configurable to:
identify, based on one or more spatial characteristics, two or more images from the plurality of images, each of the two or more images showing the one or more persons engaging in at least one action indicative of engagement with the machinery;
determine, based on temporal characteristics of the two or more images, that the at least one action is associated with a composite event and an associated time series of actions; and
determine, based on the one or more spatial characteristics, that the one or more persons are performing the composite event in behavior indicative of an unsafe practice; and
thereafter provide safety information to a visual display system, connected software system or other decision support tool;
wherein to determine that the one or more persons are performing the composite event in the behavior indicative of the unsafe practice is based on a Safety Activity Threshold (SAT);
wherein the sensor data further comprises machine state data, and wherein the SAT is contextually adjusted based on the machine state data; and
wherein the processor contextually adjusts the SAT to indicate an increased probability that the one or more persons are performing the composite event in the behavior indicative of the unsafe practice responsive to the machine state data indicating that the machinery is active.