Research On Deep Learning-based Billiards Cue Stroke Training Assistance System

With the global popularity of billiards,an increasing number of billiards enthusiasts are eager to improve their competitive level.In response to the issue of standardized cue stroke training in billiards,a deep learning-based billiards cue stroke training assistance system was designed and researched by improving key target detection algorithms and models.Firstly,a private dataset of key billiards targets was created,which includes 4861 images covering four major categories: cue ball,cue stick,bridge,and pool table,thereby addressing the current lack of billiards datasets.Secondly,based on the YOLO v5 target detection model,an improved YOLO v5 key billiards target detection model was proposed by improving the feature extraction structure,adding attention mechanisms,and improving the loss function,achieving real-time detection of key billiards targets.Finally,a deep learning-based billiards training assistance system was developed,with main functions including real-time key billiards target detection,cue stick posture data collection,cue stroke analysis,and user interaction.The system uses deep learning technology to achieve real-time detection of key targets in the billiards scene,and analyzes cue strokes in conjunction with posture sensor data,thereby assisting users in improving billiards cue stroke skills.The main contents of this paper are as follows:(1)To address the key problem of missing datasets in current billiard motion target detection.A billiard key target dataset containing four categories of mother ball,cue ball,hand frame and cue table is produced.(2)Taking into full account the significant scale differences of key targets in billiards,an improved YOLO v5 key billiards target detection model was proposed.By using Bi FPN to optimize the feature extraction structure,adding SE attention mechanisms,and improving the CIOU loss function,an improved YOLO v5 key billiards target detection model was proposed.(3)A deep learning-based billiards cue stroke training assistance system was proposed,aimed at assisting users in improving the efficiency of cue stroke training in billiards.The system uses real-time video stream collection and processing technology,combined with the improved YOLO v5 algorithm to perform real-time detection of key targets in the billiards scene.At the same time,the system collects cue stick action information through posture sensors,wirelessly transmits it to the host,and analyzes cue strokes in conjunction with target detection results.The system also provides a friendly graphical user interface,displaying real-time detection results and sensor data,and supports user login and registration functions.