基于模糊神经网络和图模型推理的动作识别

Action Recognition Based on Fuzzy Neural Network and Graph Model Inference

提出了一种基于模糊神经网络和图模型推理的人体动作识别方法。该方法将模糊神经网络和概率图模型推理有效地结合起来,目的在于能够更加准确、容易地对复杂视频动作序列进行识别,获得较高的动作识别精度。该方法由系统学习阶段和动作识别阶段两部分组成。首先,在系统学习阶段,构建了一个动态语义识别的层次图模型结构:第一级是基于模糊神经网络的关键帧动作识别图模型,用于关键帧的动作识别;第二级是关键帧序列分类模型,用于关键帧序列的语义识别。其次,在动作识别阶段,使用模糊神经网络和图模型推理算法进行动作识别,从而得到识别结果。最后,通过对实验结果的分析比较可以看出,相比现有的人体动作识别方法,该方法具有更好的识别性能,识别结果精度更佳。

We propose a novel action recognition method based on fuzzy neural network and graph model （FNNGM）. It combines the fuzzy neural network and the probability graph model reasoning effectively,in order to be able to identify the complex video action sequence more accurately and easily and obtain a higher accuracy of action recognition. The algorithm consists of system learning and action recognition. In system learning,we firstly build a hierarchical graph model for action semantic recognition：the first level is the FNN-based representative frame （RF） gesture identification graph model which is utilized to recognize RF’s gestures;the second level is RF sequence classification model which is used to identify final semantic of RF sequence. Then in action recognition,the action recognition results are computed using the FNNGM inference algorithm. Finally,through the analysis and comparison of the experimental results,it can be concluded that the proposed method has better recognition performance and accuracy than some existing methods.