System and method to predict, prevent, and mitigate workplace injuries
A prevention and safety management system utilizes a non-intrusive imaging sensor (e.g. surveillance cameras, smartphone cameras) and a computer vision system to record videos of workers not wearing sensors. The videos are analyzed using a deep machine learning algorithm to detect kinematic activities (set of predetermined body joint positions and angles) of the workers and recognizing various physical activities (walk/posture, lift, push, pull, reach, force, repetition, duration etc.). The measured kinematic variables are then parsed into metrics relevant to workplace ergonomics, such as number of repetitions, total distance travelled, range of motion, and the proportion of time in different posture categories. The information gathered by this system is fed into an ergonomic assessment system and is used to automatically populate exposure assessment tools and create risk assessments.
1. A system for assessing worker injury risks, the system comprising:
an image capturing device that simultaneously captures two-dimensional (2D) image data of a first worker and a second worker; and
a computing device in communication with the image capturing device to receive the 2D image data from the image capturing device, the computing device being adapted to:
execute a machine learning model that includes a neural network and that is applied to the 2D image data to generate, using a motion manifold and in a special orthogonal group format, a first set of joint positions of the first worker, a first set of body angles of the first worker, a second set of joint positions of the second worker, and a second set of body angles of the second worker,
execute a Monte Carlo Kalman filter to track the first worker and the second worker,
identify at least one individual worker based on the first set of joint positions, the first set of body angles, the second set of joint positions, and the second set of body angles,
parse the first set of joint positions, the first set of body angles, the second set of joint positions, and the second set of body angles into metrics relevant to workplace ergonomics, and
populate a worker risk assessment tool with the metrics relevant to workplace ergonomics to generate waveform graphs of the joint positions and body angles to predict a first worker future risk assessment of the first worker and a second worker future risk assessment of the second worker, each waveform graph including an angular flexion of a joint on one axis, an angular velocity of the joint on one axis and moments of time on a second axis.
2. The system of claim 1 , wherein the computing device is a computing cloud.
3. The system of claim 1 , wherein the metrics relevant to workplace ergonomics comprise posture categories.
4. The system of claim 3 , wherein the metrics relevant to workplace ergonomics further comprise movement velocities.
5. The system of claim 3 , wherein the metrics relevant to workplace ergonomics further comprise rest times.
6. The system of claim 3 , wherein the metrics relevant to workplace ergonomics further comprise variations in movement patterns.
7. The system of claim 1 , wherein the computing device is further adapted to recommend a course of action to address risks identified in the first worker future risk assessment and the second worker future risk assessment.
8. The system of claim 1 , wherein the computing device is further adapted to execute the machine learning model, without using data from wearable sensors, to determine ergonomic data related to posture, lift, pull, reach, force, repetition, and duration.
9. The system of claim 1 , wherein:
the first worker future risk assessment includes information about the grade and risk of future injuries to the first worker; and
the second worker future risk assessment includes information about the grade and risk of future injuries to the second worker.
10. The system of claim 1 , wherein the computing device is adapted to
execute the Monte Carlo Kalman filter to track the first worker and the second worker in frame-to-frame of the 2D image data.
11. The system of claim 1 , the computing device is further adapted to:
select at least one performance metric from a plurality of performance metrics, each performance metric from the plurality of performance metrics associated with an ergonomic movement of at least one of the first worker or the second worker, and
generate at least one performance metric weight associated to the at least one performance metric, the at least one performance metric weight indicating a relative importance of the at least one performance metric compared to the remaining performance metrics from the plurality of performance metrics.
12. The method of claim 1 , wherein the neural network of the machine learning model is a region-based convolutional neural network (R-CNN).
13. A method of assessing worker ergonomic risks, comprising:
receiving two dimensional (2D) image data from an image capturing device positioned to capture image data of one or more workers simultaneously performing tasks, the one or more workers not wearing motion tracking equipment;
executing a machine learning model that includes a neural network and that is applied to the 2D image data to generate, using a motion manifold and in a special orthogonal group format, a set of joint positions and body angles of each worker from the one or more workers;
executing a Monte Carlo Kalman filter to track the at least one individual worker;
generating a landmarks map of multiple joints of at least one individual worker from the one or more workers;
recording positions and angular relations of the joints in the landmarks map in each frame of the 2D image data;
generating a waveform graph for the at least one individual worker based on the recorded positions and angular relations; and
assessing future risks for the at least one individual worker based on an analysis of the waveform graph, the waveform graph including an angular flexion of a joint on one axis, an angular velocity of the joint on one axis and moments of time on a second axis.
14. The method according to claim 13 , wherein:
the recorded positions of the joints are in two dimensions within each frame of the 2D image data; and
the two-dimensional positions are used to estimate three dimensional positions of the joints.
15. The method of claim 13 , further comprising:
selecting at least one performance metric from a plurality of performance metrics, each performance metric from the plurality of performance metrics associated with an ergon omic movement of the at least one individual worker, and
generating at least one performance metric weight associated to the at least one performance metric, the at least one performance metric weight indicating a relative importance of the at least one performance metric compared to the remaining performance metrics from the plurality of performance metrics.