Refining Fitness Functions and Optimising Training Data in GP for Object Detection

摘要

This paper describes an approach to the refinement of a fitness function and the optimisation of training data in genetic programming for object detection particularly object localisation problems. The approach is examined and compared with an existing fitness function on three object detection problems of increasing difficulty. The results suggest that the new fitness function outperforms the old one by producing far fewer false alarms and spending much less training time and that some particular types of training examples contain most of the useful information for object detection.