An Improved Pose Matching Method Based on Deep Learning
Junyi Li, Jun Li · 2019 IEEE 1st International Conference on Civil Aviation Safety and Information Technology (ICCASIT) · 2019
An improved pose matching method based on the combination of Mask-RCNN and correlation coefficient is proposed to improve the accuracy of pose matching in images. Firstly, the feature points of two-dimensional human pose joints in the test set are correctly identified and extracted by fine-tuned Convolutional Neural Networks. Then, the joint feature points extracted for the same target in the two images are not necessarily the corresponding feature point pairs, the correlation coefficient matching algorithm is used for correction. Compared with the traditional algorithm, this method eliminates the feature points outside the target segmentation area effectively, which reduces the mismatch and improves the matching accuracy. The experimental results verify the effectiveness of the method and the practicability of the system.