Transparent Peer Review By Scholar9
Human Activity Classification From Images Using CNN
Abstract
Human activity recognition is an important as well as a challenging problem as far as computer vision is concerned. We propose inclusion of context features along with the machine learning model developed in this work, in order to determine the particular subject activity in the image. In order to improve the effectiveness of the recognition, we reuse the data from the existing dataset compiled from such sources. To present high level associated with human activity recognition concerning the collection of images, we create the architecture of the deep neural network. Human activity recognition involves prediction of people actions based on visual information. The photographs are organized according to the different types of activities that have been carried out. The objective is to use the machine learning approaches for human activity forecasting with the highest level of accuracy. It is necessary to note that the photos can be categorized into different classes using the CNN Algorithm. To make the optimal architecture choices more than two architectures comparison was made. As a last stage, the model can be implemented over the Django framework.
Aravind Ayyagari Reviewer
25 Sep 2024 02:48 PM
Approved
Relevance and Originality
The research addresses the significant challenge of human activity recognition (HAR) in computer vision, an area of growing importance with applications in various fields such as surveillance, healthcare, and smart environments. The proposal to incorporate context features alongside traditional machine learning models is a novel approach that could enhance recognition accuracy. However, further elaboration on the unique aspects of the proposed solution compared to existing methods would strengthen its originality.
Methodology
The methodology mentions the development of a deep neural network architecture for HAR but lacks specific details on the design and training process. Clarifying the data preprocessing steps, the criteria for selecting context features, and the evaluation metrics used would improve the robustness of the methodology section. Additionally, providing information about the existing datasets utilized and how they were modified for this study would be beneficial.
Validity & Reliability
To establish the validity and reliability of the findings, the paper should present detailed performance metrics, such as accuracy, precision, recall, and F1 score. Discussing the training and validation splits, potential overfitting issues, and cross-validation techniques would enhance the credibility of the results. Additionally, providing examples of how the model performs under different conditions (e.g., lighting, occlusions) would be valuable.
Clarity and Structure
The writing presents key concepts but could benefit from improved clarity and organization. Clearly defined sections—such as introduction, methodology, results, and conclusion—would aid in readability. Including visual aids like diagrams or flowcharts to illustrate the model architecture and workflow would enhance understanding.
Result Analysis
While the objective of achieving high accuracy in HAR is stated, the analysis lacks specific quantitative results or comparisons with baseline models. Including performance evaluations of the different architectures compared, along with visualizations of the results, would enrich the result analysis. Additionally, discussing the implications of the findings for practical applications and potential future work would provide valuable context and depth to the research.
IJ Publication Publisher
Done Sir
Aravind Ayyagari Reviewer