School of Art, Shandong University of Finance and Economics, Jinan 250014, China.
Comput Intell Neurosci. 2022 Jun 3;2022:2654892. doi: 10.1155/2022/2654892. eCollection 2022.
With the popularization and development of Internet of Things technology, a large number of music and dance videos have emerged in all walks of life. In this information age, video communication has become a widespread communication method. In the current music and dance collection process, most of the action frame information of the dance video is repeated, and the stage background and costumes of the dance action are too many to fully express the human body movement information. Based on these problems, this article will realize the application of the intelligent sensor-based action recognition technology in the field of dance movement collection and complete the collection and recognition of music and dance movements. The research results of the article show that: (1) in the dance video image extraction process, the feature recognition effect of the proposed algorithm is the highest among the three models. The recognition effect of the upper body is 66.1, and the recognition effect of the lower body is 61.0. The image recognition effect can reach 73.4. During the statistical experiments on the recognition of different regions of the human body, the recognition effect of the intelligent sensor model proposed in the article is still the highest among the three models. The recognition effect of the upper body is 33.9, and the recognition effect of the lower body is 33.9. The recognition effect is 34.5, and the recognition effect of the whole body is 40.7. (2) In the traditional music and dance collection mode, the values of the four test parts are all greater than 0.05, indicating that in the traditional music and dance collection mode, the differences between the four test modules are not significant. Combined with the evaluation results of the three groups in the traditional music and dance collection mode, we can conclude that under the condition that the initial conditions are basically the same, and the training conditions and environment are basically the same, the trainees who use the smart sensor music and dance collection training method are better in physical fitness. The indicators have been better improved, and the effect is greatly optimized compared with the training effect in the traditional music and dance collection mode. (3) After the test set runs, the article proposes that the accuracy rate of the dance collection model based on the smart sensor algorithm is 88.24%, the accuracy rate can reach 88.96%, the improved accuracy rate can reach 91.46%, and the accuracy rate can reach 91.79%. The ROC curve value of the article and the improved model is very stable. The ROC value before the improvement remains at about 0.90, and the ROC value after the model improvement also remains at 0.96. After the test set runs, the performance of the four models has decreased to a certain extent, but the smart sensor dance acquisition model proposed in the article has the lowest degree of decline, and the performance after the decline is still the highest among the four models. The accuracy of the model is 90.24%, and the accuracy of the improved model is 93.16%. The ROC curve values of the improved system are very stable, the ROC value has been maintained at 0.95, and the ROC value before the improvement is stable within the range of 0.85-0.95. The experimental results further illustrate that the model proposed in the article has the best performance.
随着物联网技术的普及和发展，各行各业涌现出大量的音乐和舞蹈视频。在这个信息时代，视频通信已经成为一种广泛使用的通信方式。在当前的音乐和舞蹈采集过程中，舞蹈视频的动作帧信息大部分是重复的，舞台背景和服装的舞蹈动作太多，无法充分表达人体动作信息。基于这些问题，本文将实现基于智能传感器的动作识别技术在舞蹈动作采集领域的应用，完成音乐和舞蹈动作的采集和识别。本文的研究结果表明：（1）在舞蹈视频图像提取过程中，所提出算法的特征识别效果在三种模型中最高。上身的识别效果为 66.1，下身的识别效果为 61.0。图像识别效果可达 73.4。在对人体不同区域识别的统计实验中，文章提出的智能传感器模型的识别效果仍然在三种模型中最高。上身的识别效果为 33.9，下身的识别效果为 33.9。识别效果为 34.5，全身的识别效果为 40.7。（2）在传统的音乐和舞蹈采集模式中，四个测试部分的值均大于 0.05，表明在传统的音乐和舞蹈采集模式中，四个测试模块之间的差异不显著。结合传统音乐和舞蹈采集模式下三组的评价结果，可以得出结论，在初始条件基本相同、训练条件和环境基本相同的情况下，使用智能传感器音乐和舞蹈采集训练方法的学员在身体素质方面有了更好的提高，各项指标都得到了更好的提高，与传统音乐和舞蹈采集模式的训练效果相比，效果有了很大的优化。（3）在测试集运行后，文章提出基于智能传感器算法的舞蹈采集模型的准确率为 88.24%，准确率可达 88.96%，改进后的准确率可达 91.46%，准确率可达 91.79%。文章和改进模型的 ROC 曲线值非常稳定。改进前的 ROC 值保持在 0.90 左右，改进后的 ROC 值也保持在 0.96。在测试集运行后，四个模型的性能都有一定程度的下降，但文章提出的基于智能传感器的舞蹈采集模型下降程度最低，下降后的性能仍然是四个模型中最高的。模型的准确率为 90.24%，改进后的模型准确率为 93.16%。改进系统的 ROC 曲线值非常稳定，ROC 值一直保持在 0.95，改进前的 ROC 值稳定在 0.85-0.95 范围内。实验结果进一步表明，本文提出的模型性能最佳。
