Vision-based Breast Self-Examination Hand Interaction Tracking using Sparse Optical Flow and Genetic Algorithm

摘要

Breast cancer is the leading cause of cancer mortality among women worldwide. Breast self-examination (BSE) is among the methods that can raise breast awareness, especially in developing countries where the resources are limited. However, there's currently no objective characterization of BSE performance. In this paper, we propose a feature-based BSE hand-to-breast interaction tracking method by sparse optical flow of corner points and genetic algorithm. Firstly, corner features are detected by Harris detection and a motion mask is applied to focus only on the dynamic features, which are then subjected to sparse optical flow. Then, the hand-to-breast interaction is tracked by genetic algorithm with a fitness function dependent on the number of neighbors within an arbitrary cluster radius, and magnitude/angle standard deviation values of optical flow vectors. Finally, the proposed method was verified with seven actual BSE video sequences and the result exhibited successful tracking with best accuracy of 90.2 % and an average accuracy of 83.5 %, respectively.