Action recognition is a domain that gains more and more interest, along with the development of specific motion capture hardware and processing power. Its many applications in domains such as national security and behavior analysis make it even more popular among the scientific community, especially considering the ascending trend of machine learning methods. All of these reasons have lead to a recent spike in methods and approaches necessary to solve real life problems through human actions recognition.In human actions recognition, there are mainly two approaches when attempting to build a classifier, either by using the RGB images, or by the sensor data, or where if possible a combination of these two. Both methods have their advantages and disadvantages; purely using video data makes it possible to build a classifier on almost any infrastructure, as it usually does not require specialized equipment, whereas combining it with the sensor data provides a higher accuracy, albeit at a higher cost.Both approaches have their domains of utilization in real life problems, solvable through action recognition.Neural networks and especially the convolutional neural networks are a starting point for solving the problem of human action recognition. Due to their nature, they can recognize spatial and temporal features very well, making them ideal for working with RGB images or, in this case, sequences of RGB images.In the present paper is proposed a convolutional neural network architectures based on 2D kernels. It is presented its structure, advantages and disadvantages, along with metrics that measure the performance. This solution, based on 2D convolutions, is the fastest, but with the lowest performances compared with other known solutions. The main problem when dealing with videos is extracting the context from a sequence of frames. Trying to be classified the video using 2D Convolutional Layers, which may be translated to classifying an example by either the most significant frame, or frame by frame, by applying a probability distribution over the partial classes in order to obtain the final prediction. It proves to be very simple to differentiate between actions, especially since the difference between them is subtle and consists of only a small part of the overall image. When classifying via the key frames, the total accuracy obtained is around 10%. The other approach, classifying each frame individually, proved to be too computationally expensive with negligible gains.Next some obtained results are mentioned. The paper ends with some conclusions and references.

Model of Human Action Recognition Based on 2d Kernel

Paper Proceedings: https://asmedigitalcollection.asme.org/ISPS/ISPS2021/volume/84799

ASME 2021 Information Storage and Processing Systems (ISPS2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.