Genetic search for fast object detection

Many practical applications requires fast and robust object detection as their first step. However, most existing robust methods are computationally expensive. This paper describes a new object detection framework to reduce computational cost while retaining high detection accuracy and robustness. In the framework, a genetic algorithm (GA) is used to search an input image efficiently while a neural network (NN) serves as an object filter. Each individual in the GA represents a subwindow extracted from the image. The individuals are given fitness according to how well they match the NN-based object filter. Based on their fitness, the genetic search is guided to possible object areas. Experiments in the domain of face detection are presented and the results show the effectiveness of the proposed method.