DefinePK hosts the largest index of Pakistani journals, research articles, news headlines, and videos. It also offers chapter-level book search.
Title: Codebook-Based Feature Engineering for Human Activity Recognition Using Multimodal Sensory Data
Authors: Seerat Fatima, Laiba Zahid, Tazeem Haider, Muhammad Hassan Khan, Muhammad Shahid Farid
Journal: International Journal of Innovations in Science & Technology
Publisher: 50SEA JOURNALS (SMC-PRIVATE) LIMITED
Country: Pakistan
Year: 2024
Volume: 6
Issue: 7
Language: English
Keywords: Multimodal Sensory DataCodebookBag of FeaturesMini Batch K-MeansSoft Assignment.
Recently, Human Activity Recognition (HAR) using sensory data from various devices has become increasingly vital in fields like healthcare, elderly care, and smart home systems. However, many existing HAR systems face challenges such as high computational demands or the need for large datasets. This paper introduces a codebook-based approach designed to overcome these challenges by offering a more efficient method for HAR with reduced computational costs. Initially, the raw time series data is segmented into smaller subsequences, and codebooks are constructed using the Bag of Features (BOF) approach. Each subsequence is then assigned softly based on the center of each cluster (codeword), resulting in a histogram-based feature vector. These encoded feature vectors are subsequently classified using a Support Vector Machine (SVM). The proposed method was evaluated using the OPPORTUNITY dataset, comprising data from 72 sensors, achieving a classification accuracy of 90.7%. In comparison to other advanced techniques, our approach not only demonstrated superior accuracy in recognizing human activities but also significantly reduced computational costs. The use of soft assignments for mapping codewords to subsequences efficiently captured the key patterns within the activity data. The findings validate that the proposed codebook-based method provides substantial improvements in both accuracy and efficiency for HAR systems.
To develop a computationally efficient method for Human Activity Recognition (HAR) using sensory data through a codebook-based approach, aiming to improve accuracy while reducing computational requirements.
The proposed method involves segmenting raw time-series data into subsequences, constructing codebooks using the Bag of Features (BOF) approach with Mini Batch K-Means clustering, and then performing feature encoding using soft assignments to create histogram-based feature vectors. These vectors are subsequently classified using a Support Vector Machine (SVM) with a linear kernel. The method was evaluated on the OPPORTUNITY dataset.
graph TD
A[Load Dataset] --> B[Data Preprocessing: Cleaning, Normalization];
B --> C[Data Segmentation: Sliding Window];
C --> D[Codebook Construction: Mini Batch K-Means];
D --> E[Codeword Extraction];
E --> F[Feature Encoding: Soft Assignment];
F --> G[Histogram-based Feature Vector Generation];
G --> H[Split Data: Training/Testing];
H --> I[Train SVM Classifier];
I --> J[Predict Labels for Test Data];
J --> K[Evaluate Model Accuracy];
The codebook-based approach, particularly with soft assignment, effectively captures subtle variations in human activities, leading to high accuracy and efficiency. This method offers a more lightweight and scalable solution than deep learning models, though further testing on diverse datasets is recommended.
The proposed codebook-based method achieved a classification accuracy of 90.7% on the OPPORTUNITY dataset, demonstrating superior accuracy and significantly reduced computational costs compared to other advanced techniques.
The paper presents an efficient codebook-based framework for HAR that operates effectively without requiring large datasets or significant computational resources. The method, evaluated on the OPPORTUNITY dataset, achieved an impressive accuracy of 90.7%, outperforming existing state-of-the-art techniques.
1. Accuracy: The paper reports a classification accuracy of 90.7% on the OPPORTUNITY dataset. This is confirmed in Table 3 and the Results section.
2. Dataset: The study utilized the OPPORTUNITY dataset, which contains data from 72 sensors across 10 modalities. This is confirmed in the "Dataset" subsection of the Results and Discussion.
3. Year of Publication: The paper is published in the International Journal of Innovations in Science & Technology, Special Issue, October 2024. This is indicated on the page headers and citation information.
Loading PDF...
Loading Statistics...