Detection of Principal Component Features for Person Identification in Video

Abstract

In the field of pattern recognition, computer vision and biometrics, various applications are found and brought into real-time application use. Person Detection system made possible to the modern world to achieve better and faster surveillance in crowded areas as it is impossible for human. Person Detection Model consists mainly of three stages namely: Pre-processing, Feature Extraction and Classification. This work explored with Subspace techniques, like Principal Component Analysis (PCA) and Fisher Linear Discriminant Analysis (FLDA) as feature extractors. Learning Vector Quantizer (LVQ) was used as multi-layer network classifier. 100% Recognition rate was obtained using PCA and FLDA through LVQ classifier for Visual Tracker Benchmark Database. 91.67% Recognition rate was obtained using PCA with LVQ classifier for Home Video database. Many challenges were addressed in this research such as variation in light, pose variations (>20 degree) and variation of facial expressions in video. Two minutes of minimal Execution time is required. 100% accuracy was obtained considering 1000 epochs.