Video target detection method based on I-P frame feature fusion

The invention discloses a video target detection method based on I-P frame feature fusion, and the method comprises the steps: determining sample data and evaluation indexes used for network training, carrying out the data enhancement of the sample data before the training, transmitting the sample data after the data enhancement processing into an IP-Net network structure for training, and carrying out the training of the IP-Net network structure; the training process comprises the steps of performing feature extraction on key frames in a video by adopting an I-Net network model to obtain key features, and performing feature extraction on prediction frames by adopting a P-Net network model before the next key frame to obtain prediction features; and carrying out feature fusion on the key features and the prediction features, and taking the fused features as a final output result of the IP-Net network. According to the method, the complex model network I-Net and the simple model network P-Net are used for cros