Swimming action recognition algorithm based on improved C3D and attention-residual network

The traditional swimming analysis method has the disadvantages of strong subjectivity, difficult quantification, high cost, easy to interfere with training, and the existing three-dimensional convolution network is affected by water ripples and bubbles in the underwater scene, and the key feature extraction is insufficient.A swimming motion recognition algorithm based on improved three-dimensional convolution and attention residual network is proposed.The experimental results show that, compared with the baseline C3d model, the number of parameters of the algorithm is reduced by 42%, the floating-point operation is reduced by 58%, the reasoning speed is reduced by 61%, and the recognition accuracy is improved by 8.3% compared with the baseline C3d model under the same hardware configuration.This method provides a feasible and efficient new way for the quantification, real-time feedback and intelligent teaching of swimming technology.