An Evolutionary Computing Approach to Solve Object Identification Problem for Fall Detection in Computer Vision-Based Video Surveillance Applications

Abstract

This chapter proposes to present the design and development of an evolutionary computation (EC) based system which solves two interesting real-world problems. In the first phase of the system, the object detection problem of image processing is considered. An EC framework is designed to solve it, by formulating it as an optimization problem, considering the minimal required features to detect an object in a given image. The differential evolution (DE) algorithm is used in the framework. The objectives considered are to maximize the accuracy of detection and minimize the number of features required. Belga LOGOS dataset is considered, and 10 different feature descriptors are individually applied for object detection. Later applied the combination of features using the proposed EC framework and compared the results. Individual descriptors gave an average accuracy of 72.35% whereas the proposed method gave an average accuracy of 81.15% which shows the efficiency of the proposed framework. In the second phase, a real-world problem of fall detection of elderly people is considered for surveillance applications. A vision-based solution is proposed instead of using any external devices such as sensors and accelerometers to detect the fall. The advantage of this method is that it doesn’t need the subject to carry any devices and is cost-effective as it is based on only the surveillance videos. To detect the fall, the person needs to detected and tracked during their activities. Initially, people are tracked using core computer vision technique later applied the EC framework developed in Phase I to detect the people and the results are compared. Core computer vision techniques gave an accuracy of 97.16% whereas the proposed optimization framework gave an accuracy of 98.65%. This proved the efficiency of the proposed approach.