The model based on 2D convolutions, is nowadays the fastest model, but with the lowest performances. The model based on 3D Convolutions is a middle ground, useful in certain situations which require a fast classifier operating on different action classes.Behavior analysis is a domain greatly improved by action recognition. By placing human actions in different categories, is possible to extract statistics regarding a person's behavior, which can then be processed by specialized personnel, depending on the problem at hand. For instance, in a nursing home, extracting patterns of its residents' routine can determine if a person might need additional or specialized care. Moreover, when a specific action committed by a resident is deemed dangerous, i.e. falling, walking without aid, getting up from bed, etc., a nurse could be alerted in order to limit the possibly grave consequences. In the nursing home case, this solution provides a fail-safe panic button for the residents who either cannot or refuse to use other specialized devices.Action recognition infrastructure can be applied anywhere behavior analysis is required. Since a video approach can be built over an already existing infrastructure comprised of CCTVs (closed-circuit television) and a central computer, it can be used in any space where monitoring certain actions is needed.This model has good results, with an accuracy of approximately 55%, since it takes into account the context between frames. It is presented the Optimizer and Loss Function, associated with the general structure and the proposed architectures. Loss functions compute a measure of how well the weights of the network's parameters perform in the task, or a measure of the distance between the output of the network and the absolute value. Optimizers have the role of updating the problems of the network, with respect to the loss function and the network structure. In computer vision there are multiple combinations of loss functions and optimizers, but in this paper is considered the Cross Entropy Loss and Adam Optimizer. Finally, are mentioned some obtained results, conclusions and references.

Modeling 3D Convolution Architecture for Actions Recognition

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.