Model transfer from 2D to 3D study for boxing pose estimation.

Introduction: Boxing as a sport is growing on Chinese campuses, resulting in a coaching shortage. The human pose estimation technology can be employed to estimate boxing poses and teach interns to relieve the shortage. Currently, 3D cameras can provide more depth information than 2D cameras. It can potentially improve the estimation. However, the input channels are inconsistent between 2D and 3D images, and there is a lack of detailed analysis about the key point location, which indicates the network design for improving the human pose estimation technology. Method: Therefore, a model transfer with channel patching was implemented to solve the problems of channel inconsistency. The differences between the key points were analyzed. Three popular and highly structured 2D models of OpenPose (OP), stacked Hourglass (HG), and High Resolution (HR) networks were employed. Ways of reusing RGB channels were investigated to fill up the depth channel. Then, their performances were investigated to find out the limitations of each network structure. Results and discussion: The results show that model transfer learning by the mean way of RGB channels patching the lacking channel can improve the average accuracies of pose key points from 1 to 20% than without transfer. 3D accuracies are 0.3 to 0.5% higher than 2D baselines. The stacked structure of the network shows better on hip and knee points than the parallel structure, although the parallel design shows much better on the residue points. As a result, the model transfer can practically fulfill boxing pose estimation from 2D to 3D.

Method for Identifying Maize Haploid Seeds by Applying Diffuse Transmission Near-Infrared Spectroscopy.

The identification for haploid seeds is an important process in maize haploid breeding. Thanks to the diffuse transmission (DT) technology of near-infrared (NIR) spectroscopy, maize haploid seeds can be selected automatically using NIR spectrum features. However, the NIR spectra of maize seeds contain a large number of redundant features and noise that will degrade the identification performance. We resolved this problem by designing a low dimension and uniform space of seed spectrum features to improve the collected spectra. The zero-phase component analysis (ZCA) method was utilized to uniform the feature space and the partial least squares regression (PLSR) was employed to design the low dimension space. Then, by using the classifier of back propagation neural network (BPNN), a high qualitative identification method was developed for selecting maize haploid seeds. The study results demonstrate that the average accuracy of the proposed method is outstanding (96.16%) with a minor standard deviation (SD) compared with other methods. Therefore, our proposed method is potentially useful for automatically identifying maize haploid seeds.