Multi-scale Spatio-temporal Feature Fusion Module Based Posture Relative Aesthetic Assessment for AI Camera

Photography is the art of creating durable images by capturing and recording light. AI cameras incorporate artificial intelligence technology to enhance the imaging functionality. The existing research has realized “smile capture” based on facial smile detection. When turning to full shots, it is necessary to analyze all movements of the subject to capture a beautiful posture. The primary problem is how to define the aesthetic standard of human posture. In this paper, the stage pose is defined as a beautiful posture in performance videos, and then an aesthetic pose dataset is built. Secondly, this paper proposed a beautiful posture-determining model based on relative aesthetic assessment, in which the multi-scale spatial features are extracted to learn fine-grained global information, and the multi-scale temporal features are combined to perceive both short-term and long-term motion information in videos, thereby enhancing the learning of fine movements in videos. The experimental results show that the proposed model reaches an accuracy of 80.79% in judging the beauty of human posture.