Face Recognition Using Genetic Algorithm - A Unique Approach With Pre-Trained Models For Low Power Microcontrollers

This paper proposes a unique approach for recognizing faces with pre-trained models using Genetic Algorithm and 2D Fast Fourier Transform in microcontrollers. Conventional real-time face recognition algorithms on low-end hardware have three limitations: a) complex algorithms are difficult to implement; b) recognizing new faces requires retraining which in-turn requires a lot of processing power, RAM and time; and c) pre-trained models (using architectures such as Neural Networks) need a lot of memory. These factors limit its usage in low-end devices where training time is restricted and the face database size is variable. Hence, most of the offline machine learning algorithms in microcontrollers are limited to text and character recognition. We propose a robust and memory-efficient system that works with pre-trained models and does not require retraining for recognizing new faces. The trained data is in the order of a few kilo-bits and the feature vector size for each image is in the order of a few kilobytes. This enables it to run on low-end Microcontrollers (MCUs) and Node MCUs. This algorithm has been evaluated using standard face datasets. Tests with these datasets show that the proposed solution using Genetic algorithm performs as good or better than other widely available techniques.