Real-time foreground detection approach based on adaptive ensemble learning with arbitrary algorithms for changing environments

- Monitoring-ensemble-learning detection (MELD) is proposed for foreground detection.
- Foreground detection technologies are integrated based on practical applications.
- Monitoring and learning mechanisms are designed for complex scenes.
- Parameters and weights for optimal parameterization are calculated automatically.
- Modular architecture is used to add, update, and remove technologies from MELD.

Foreground detection technologies have emerged as an important research area with increasing popularity of computer vision and camera devices. Even though several foreground detection approaches have been proposed, they cannot address various challenges in actual complex scenes owing to their applicability and restrictions. This study proposes a method that can integrate arbitrary detection technologies to detect foregrounds in real time, thereby improving overall detection performance of video-based systems. Moreover, the proposed approach can be fully initialized with initial foreground results, requires no training, and performs dynamic adjustments online, for every new frame. In this approach, critical weighted values are automatically calculated over time based on observed scenes for optimal flexibility and parameterization. Thus, the proposed method has the flexibility to accommodate any new technology to overcome the challenging problems of foreground detection in changing environments. Experimental results demonstrate that the performance of the proposed method is comparable to that of state-of-the-art methods and satisfies the requirements of real-time practical applications.

Graphical Abstract176