Safety vision device, and safety vision system

A safety vision device comprising:

• a human three-dimensional skeleton estimation model as a learned model based on a supervised learning receiving input of a two-dimensional image of a human, and outputting three-dimensional joint point data indicating three-dimensional coordinate values of positions of joint points of the human;
• a two-dimensional skeleton estimation model as a learned model based on a supervised learning receiving input of a two-dimensional image of a robot captured by a camera and outputting a two-dimensional posture of pixel coordinates indicating positions of the centers of the joint axes of the robot in the two-dimensional image when the two-dimensional image is captured;
• a robot three-dimensional skeleton estimation model as a learned model based on a supervised learning receiving input of the two-dimensional posture of pixel coordinates of the robot, and a distance and a tilt between the camera that has captured the two-dimensional image of the robot and the robot, and outputting angles of a plurality of joint axes included in the robot;
• an input unit configured to input a two-dimensional image of a worker and the robot captured by an external camera, and a distance and a tilt between the external camera and the robot;
• an estimation unit configured to input the two-dimensional image of the worker which have been inputted by the input unit to the human three-dimensional skeleton estimation model and estimate the three-dimensional joint point data indicating three-dimensional coordinate values of positions of joint points of the worker,
• the estimation unit further configured to input the two-dimensional image of a robot captured by the camera to the two-dimensional skeleton estimation model and estimate a two-dimensional posture of pixel coordinates indicating positions of the centers of the joint axes of the robot in the two-dimensional image,
• the estimation unit further configured to input the distance and tilt between the external camera and the robot, which have been inputted by the input unit, and the two-dimensional posture of pixel coordinates indicating positions of the centers of the joint axes of the robot to the robot three-dimensional skeleton estimation model, and estimate the angles of the plurality of joint axes included in the robot;
• and
• an approach determination unit configured to calculate an area indicating a range of the worker by adding a shape having a length, depth, and height set in advance for each straight line of the skeleton generated based on the three-dimensional joint point data of the worker, and an area indicating a range of the robot by adding a shape having a length, depth, and height set in advance to each link of the skeleton generated based on the three-dimensional joint point data and the angles of the plurality of joint axes,
• and output an instruction to decelerate or stop the robot according to a degree of overlap between the calculated areas of the worker and the robot.