Recognition of human action for scene understanding using world cup optimization and transfer learning approach

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Human activity recognition is crucial for visual scene understanding.
• Applications span surveillance, healthcare, and entertainment.
• Complex human activities involve intricate object interactions.

Purpose of the Study:

• To classify diverse human activities within visual scenes.
• To enhance the accuracy of human activity recognition models.
• To investigate the efficacy of deep learning with optimized feature selection.

Main Methods:

• Utilized pre-trained deep Convolutional Neural Networks (CNNs) like AlexNet, SqueezeNet, ResNet, and DenseNet for feature extraction.
• Employed the adaptive World Cup Optimization (WCO) algorithm for superior dominant feature selection.
• Classified features using the DenseNet 201 fully connected classifier.
• Pre-processed datasets with fuzzy color stacking.

Main Results:

• Achieved a classification accuracy of 94.7% using DenseNet for feature extraction and WCO for selection.
• Demonstrated superior performance compared to models without feature selection.
• The proposed methodology with double filtering via WCO improved classification model quality.

Conclusions:

• The integration of deep CNNs with WCO-based feature selection significantly enhances human activity recognition.
• The adaptive WCO algorithm effectively identifies dominant features, improving classification accuracy.
• This approach offers a robust solution for visual scene understanding and activity classification.