The progress in artificial intelligence has resulted in the development of highly authentic AI-generated videos. This enables malicious individuals to effortlessly produce non-existent videos and disseminate them across the internet. This paper proposes a fake video detection method, Deepfake Detection Model based on Optical Flow Estimation(DFDOF), which is based on analyzing the speed of objects in videos. Specifically, a ResNet subdetector is employed to extract image features, a multi-scale optical flow feature extraction module is used to extract optical flow features, and a Co-Attention module is utilized to fuse the features from both branches. Finally, a transformer layer is applied to classify the fused features, further enhancing the discrimination capability. Extensive experimental results demonstrate the strong generalization ability and robustness of our method.