Method, device, and non-transitory computer-readable recording medium for estimating information on golf swing

A method performed in a device for estimating information on a golf swing, the device comprising one or more processors and the method comprising the steps of:

• by the one or more processors, light-weighting an artificial neural network model using depthwise convolution and pointwise convolution to obtain a light-weighted artificial neural network model;
• by the one or more processors, when a photographed image of a user's golf swing is acquired, detecting at least one joint of the user from the photographed image using the light-weighted artificial neural network model;
• by the one or more processors, estimating a posture of the user with reference to at least one of a type of the at least one joint of the user, a position of the at least one joint of the user, a distance between the at least one joint of the user and at least one other joint of the user, and an angle formed between the at least one joint of the user and at least one other joint of the user;
• by the one or more processors, comparing the user's golf swing and a golf swing of a comparison target with reference to the posture of the user and at least one comparison point;
• and
• by the one or more processors, estimating information on the user's golf swing on the basis of a result of the comparison,
• wherein the at least one comparison point includes at least one of the position of the at least one joint, a position of a specific body part of the user estimated from the at least one joint, a reference line formed from the position of the at least one joint, and an angle formed from two or more reference lines,
• wherein the at least one comparison point is established separately for each partial motion constituting each of the user's golf swing and the comparison target's golf swing, and established separately for each point of view with respect to the same partial motion,
• wherein the user's golf swing and the comparison target's golf swing are constituted by a plurality of stages of partial motions, respectively, and similarity between the user's golf swing and the comparison target's golf swing is identified in consideration of the plurality of stages of partial motions, and
• wherein the method further comprises:
• extracting at least one frame corresponding to each stage of the plurality of stages of the user's golf swing from photographed images of the user's golf swing;
• determining a comparison point for the extracted at least one frame corresponding to each stage of the plurality of stages of the user's golf swing, wherein the comparison point is varied depending on the stage to which the extracted at least one frame corresponds, and for each stage of the plurality stages of the user's golf swing, the comparison point is determined separately for each point of view including a front view and a side view of the user's golf swing;
• and
• for each stage of the plurality of stages of the user's golf swing, comparing the comparison point for the extracted at least one frame corresponding to the stage of the user's golf swing to a target comparison point for a frame corresponding to a stage among the plurality of stages of the comparison target's golf swing corresponding to the stage of the user's golf swing, and
• wherein the plurality of stages include an address stage, a takeaway stage, a back swing stage, a top-of-swing stage, a down swing stage, an impact stage, a follow-through stage, and a finish stage.