Infant dangerous action detection and identification method based on pose estimation and improved YOLOXtiny

• Main Authors: ZHU ZEDE, WANG QISHENG
• Format: Patent
• Language: Chinese; English
• Published: 09.05.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses an infant dangerous action detection and identification method based on pose estimation and improved YOLOXtiny, and the method comprises the steps: obtaining a video frame from a camera, and carrying out the preprocessing of the input video frame; secondly, inputting the preprocessed video frames into a position and attitude estimation algorithm and a YOLOXtiny target detection algorithm improved based on model pruning and a fusion space attention mechanism so as to obtain the number of skeleton points, interested target areas and category detection result data in each frame of picture; wherein the target detection algorithm is obtained by using a self-built data set and training in a transfer learning mode; then, the obtained data are processed and fused, so that the accurate recognition capability of different dangerous actions is improved; and finally, designing rules by utilizing the processed and fused data to judge normal and dangerous action states of the infant and give an alar