Training system of negative feedback NFTS
Application.
It is used in shooting training complexes as an autonomous shooting tool that independently makes decisions to take a shot. Targeting and decision-making are performed using computer vision algorithms with a pre-trained neural model.
Criteria.
The main criterion for making a shooting decision is the assessment of the shooter's (trainee's) position. The position assessment is a task that involves determining the location of certain points in the image, known as key points.
Key points may represent various parts of the human body, such as joints, limbs, or other distinguishing features.
The result of the position assessment model is a set of points that represent the key points of the object in the image, along with reliability scores for each point.
Position assessment is the selection when you need to identify specific parts of the object and their location relative to each other.
Operating mode.
When the system determines the position of the shooter (trainee) as the person shooting, the system starts a countdown timer set by the operator, from 1 to 15 seconds.
After the time expires, if the shooter (trainee) has not changed their location, the system determines the center of the object along the X-axis, performs aiming, and takes the shot. If the shooter (trainee) has changed their location before the time set by the operator, the countdown is canceled, and the system enters standby mode.
Technical solution.
For training the position assessment model, the base model yolov8-pose was used, along with a video file of shooters and a pre-generated file with control points. Training was conducted on an NVIDIA graphics processor using Python and the Ultralytics library. To use the trained model on an NVIDIA Jetson microcomputer or on a computer with an NVIDIA graphics card, the #YOLO model file was converted to a tensorrt file. Inference is performed using the #OpenCV, #CUDA C++ libraries.
It is used in shooting training complexes as an autonomous shooting tool that independently makes decisions to take a shot. Targeting and decision-making are performed using computer vision algorithms with a pre-trained neural model.
Criteria.
The main criterion for making a shooting decision is the assessment of the shooter's (trainee's) position. The position assessment is a task that involves determining the location of certain points in the image, known as key points.
Key points may represent various parts of the human body, such as joints, limbs, or other distinguishing features.
The result of the position assessment model is a set of points that represent the key points of the object in the image, along with reliability scores for each point.
Position assessment is the selection when you need to identify specific parts of the object and their location relative to each other.
Operating mode.
When the system determines the position of the shooter (trainee) as the person shooting, the system starts a countdown timer set by the operator, from 1 to 15 seconds.
After the time expires, if the shooter (trainee) has not changed their location, the system determines the center of the object along the X-axis, performs aiming, and takes the shot. If the shooter (trainee) has changed their location before the time set by the operator, the countdown is canceled, and the system enters standby mode.
Technical solution.
For training the position assessment model, the base model yolov8-pose was used, along with a video file of shooters and a pre-generated file with control points. Training was conducted on an NVIDIA graphics processor using Python and the Ultralytics library. To use the trained model on an NVIDIA Jetson microcomputer or on a computer with an NVIDIA graphics card, the #YOLO model file was converted to a tensorrt file. Inference is performed using the #OpenCV, #CUDA C++ libraries.
Kyiv 
A little busy