A Human Posture Estimation Method for Image Interaction System Based on ECA

Nowadays, human posture estimation intelligent algorithm is widely used in image interaction system for better interaction and accuracy. In the paper we proposed an effective human posture estimation method based on multi-scale feature extraction and Efficient Channel Attention (ECA) mechanism. In our proposed method, we first extract muti-scale feature through attention-driven convolutional neural network. Besides, we use a combination of bottom-up and top-down approaches to aggregate the feature maps from different levels through multiple paths for cross-scale feature representations. Finally, the feature is regressed and classified to predict the position and confidence of the human bounding boxes, as well as the coordinates of the 17 human posture keypoints within each bounding box. Based on the human posture estimation module, we designed an image interaction system to achieve interaction between the human and the screen with action recognition and virtual scene construction. Experimental results showed that the human posture estimation method has high accuracy and real-time performance and the image interactive system has good interactivity and creativity.