The best angle correction of basketball shooting based on the fusion of time series features and dual CNN
The best shooting angle correction of basketball based on intelligent image analysis is an important branch of the development of intelligent sports. However, the current method is limited by the variability of the shape base, ignoring dynamic features and visual information, and there are some prob...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2024-12-01
|
| Series: | Egyptian Informatics Journal |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1110866524001427 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The best shooting angle correction of basketball based on intelligent image analysis is an important branch of the development of intelligent sports. However, the current method is limited by the variability of the shape base, ignoring dynamic features and visual information, and there are some problems in the process of feature extraction and correction of related actions. This paper proposes a method to correct the best shooting angle of basketball based on the fusion of time series characteristics and dual CNN. Segmenting the shooting video, taking the video frame as the input of the key node extraction network of the shooting action, obtaining the video frame with the sequence information of the bone points, extracting the continuous T-frame video stack from it, and inputting it into the spatial context feature extraction network in the shooting posture prediction model based on dual stream CNN (MobileNet V3 network with multi-channel attention mechanism fusion module), extract the space context features of shooting posture; The superimposed optical flow graph of continuous video frames containing sequence information of bone points is input into the time convolution network (combined with Bi-LSTM network of multi-channel attention mechanism fusion module), extract the skeleton temporal sequence features during the shooting movement, using the spatial context features and skeleton temporal sequence features extracted from the feature fusion module, and realizing the prediction of shooting posture through Softmax according to the fusion results, calculate the shooting release speed under this attitude, solve the shooting release angle, and complete the correction of the best shooting release angle by comparing with the set conditions. The experimental results show that this method can achieve the best shooting angle correction, and the training learning rate is 0.2 × 10–3, training loss is about 0.05; MPJPE and MPJVE indicators are the lowest, and Top-1 indicators are the highest; The shooting percentage is about 95 %. |
|---|---|
| ISSN: | 1110-8665 |