This was an assignment in one of my courses in the last semester (Spring 2020). From the problem statement:
The goal of this Assignment is to use Principal Components Analysis (PCA), K-Nearest Neighbours (kNN) and Random Forest to recognize face images and gait signals.
I experimented with some more features and classifiers along with the ones mentioned in our problem description. Find the detailed report that I prepared for the assignment in FinalReport.pdf.
I will not publish the data given to us as I don't have the permission to do so. However, I am sharing the codes and the results.
- Face Dataset: Face images of 10 people, with each person captured under 24 different lighting conditions, for a total of 240 images. These face images taken from the CMU PIE database.
- Gait Dataset: This is a gait dataset collected using Inertial Measurement Unit (IMU) sensors, with accelerometer data for the axes x, y and z. For each person, the dataset provides a csv file of acceleration values for x, y, z read at a frequency of 100Hz. These gait signals collected by NUS researchers.
- PCA: Principle Component Analysis.
- LBP: Local Binary Pattern.
- SIFT: Scale-Invariant Feature Transform.
- SURF: Speeded Up Robust Features.
- CNN: Convolutional Neural Network.
- Statistical Features (e.g. Mean, Std. deviation, Variance, etc.)
- LSTM Features.
- kNN: k-Nearest Neighbors.
- Random Forest.
- Support Vector Machine.
The fusion is a score-based fusion. The parameter (α) controls how much importance to give to each of the scores. From the problem statement:
The Final prediction from fusion is calculated by using the following formula,