Video Action Recognition based Human Behavioral Analysis using Deep 3D and R(2+1)D Convolutional Neural Networks

With the rise in the volume of multimedia content accessible on the internet, video analysis has become essential for numerous applications, including video retrieval, gaming, and behavioral analysis. Action recognition, a crucial aspect of video analysis, entails comprehending human actions within videos. In this paper, we examine the recent and efficient models for action recognition using 3D convolutional networks. Specifically, we assess the performance of the deep 3D Convolutional Neural Network (C3D) and the R(2+1)D Convolutional Neural Network in recognizing actions with a failure video dataset. Both models exhibit similar levels of accuracy, although the $\mathbf{R}(2+1) \mathbf{D}$ model demonstrates a slightly higher accuracy compared to C3D. Notably, our findings reveal that the $\mathbf{R}(\mathbf{2}+\mathbf{1}) \mathbf{D}$ models yield significantly high probabilities in predicting class action labels precisely.