Human action recognition method based on conformal geometric algebra and recurrent neural network

Abstract

Deep Learning (DL) plays an important role in machine learning and artificial intelligence. DL is widely applied in many fields with high dimensional data, including natural language processing, image recognition. High dimensional data can lead to problems in machine learning such as overfitting, degradation of accuracy. To address these issues, some methods, such as Principal Components Analysis (PCA), principal component regression (PCR), Multi-class Linear Discriminant Analysis (MLDA), were proposed to reduce dimensions of the data and computational complexity simultaneously. The drawback of these methods is that they only work well on data distributed on the plane. In the case of the data distributed on the hyper-sphere, such as objects moving in space, the processing results are not so good as expected. In this paper, we propose the use of Conformal Geometric Algebra (CGA) to extract features and simultaneously reduce the dimensionality of a dataset for human activity recognition using Recurrent Neural Network (RNN). First, human activity data in a 3-dimensional coordinate system is pre-processed and normalized by calculating deviations from the mean coordinate. Next, the data is transformed to vectors in CGA space and its dimensions are reduced to return the feature vectors. Finally, we use the RNN model to train feature vectors. Empirical results performed on the CMU eight actions dataset show that the CGA combined with RNN gave the best test results of 92.5%.

Jumping Action Recognition for Figure Skating Video in IoT Using Improved Deep Reinforcement Learning

“…When the number of experiments reaches 60, the true positive rate of different algorithms reaches the maximum value. The true positive rate of algorithms in the paper is 0.98, which is 0.23, 0.26, 0.24, 0.32, 0.29 higher than the algorithms in BRMFSJ [10], POSL [1], AHDM [16], ARMHJ [22] and HARM [13], respectively, It shows that compared with the experimental comparison algorithm, the true positive rate of the algorithm in this paper is higher, which proves that the recognition result of this algorithm is more accurate.…”

“…After the implementation of the confidence fusion scheme, the final jumping action recognition result from figure skating videos is obtained. [10], POSL [1], AHDM [16], ARMHJ [22] and HARM [13] are used for jumping action recognition, and then the true positive rate is calculated according to the action recognition results.…”

“…The features are input to improve the deep reinforcement learning network model, and the jump action recognition results are obtained. At the same time, the algorithms in BRMFSJ [10], POSL [1], AHDM [16], ARMHJ [22] and HARM [13] are applied to perform action recognition on UCF 101 dataset. According to the recognition accuracy and recall of different recognition algorithms, the PR curve image is drawn, and then the area under the curve is calculated by 11 point interpolation method to obtain the AP value.…”

“…Therefore, it is difficult to be applied to practice. In Nguyen et al [13], based on the action video, the human action data in the three-dimensional coordinate system were preprocessed and normalized by calculating the deviation from the average coordinate. The data was transformed into vectors in conformal geometric algebraic space, and its dimension was reduced to return eigenvectors.…”

Current survey of Clifford geometric algebra applications

“…The paper of Nguyen et al 130 uses CGA in order to extract features and simultaneously reduce the dimensionality of a data set for human activity recognition using a Recurrent NN. Human activity data in three dimensions are pre‐processed and normalized by calculating deviations from mean coordinates.…”