| Summary: | Computer vision, which is used to detect and track a specific target in image sequences, has drawn great attention in recent years. The process of tracking can be formulated as a dynamic optimization problem that identifies the optimal position of the target in each image. Differential evolution (DE), who owns the advantages of simplicity, parallel computing, and self-adaptive search for global optimization, is envisioned as a promising algorithm to provide effective target tracking. In this paper, several improvements are made in DE for better adaptability in target tracking. Specifically, we introduce two inferior individuals into the mutation stage, which further enriches the diversity of the population and speeds up the offspring's evolution. We also proceed several image preprocessing and build an adaptive Gaussian mixture model of the target to deal with the complex tracking scenarios. Experimental results show that the designed tracking algorithm based on our improved DE demonstrates a higher tracking accuracy and faster tracking speed in several challenging tracking scenarios.
|
|---|