The invention discloses a baby abnormal behavior detection method based on a condition
generative adversarial network and an SVM, and belongs to the technical field of
video image processing and deeplearning. Whether baby behaviors are abnormal or not is judged by analyzing baby motion trajectories in videos, firstly, baby videos are obtained, the baby videos are
cut with reasonable length and converted into frame images, and the four limbs and the
whole body are marked so that a sample
library can be built; then, the condition
generative adversarial network is used for performing target tracking on the
whole body and the four limbs of the baby; then,
wavelet approximation waveform and
wavelet power spectrum calculation is performed on the obtained target motion trajectory, obtained characteristics are classified through the SVM, and comprehensive judging is performed; motion trajectory detection is performed on information of the
whole body and the four limbs of the baby, informationis more comprehensive compared with single limb detection,
wavelet region and power spectrum combined training is adopted, the detection precision is improved, whether baby behaviors are abnormal ornot is detected, intervention is performed as soon as possible, and the method is of a great significance for preventing diseases such as baby brain
paralysis.