Research on object tracking algorithm via adaptive multi-feature fusion

Abstract

Aiming at the existing problems of object tracking in real scenes, such as complex background, illumination changes, fast motion, and object rotation, the paper has proposed an object tracking algorithm via adaptive multi-feature fusion. By extracting the HOG feature of the object and using convolutional neural networks to extract high-level and low-level convolutional features, an adaptive threshold segmentation method has been used to evaluate the effect of each feature, and the weight ratio of feature fusion has been obtained. The response map of each feature has fused according to the weight coefficient, and the new estimated position of the object has been obtained, and the object scale has been calculated by the scale correlation filter, and the object scale has been obtained to complete the object tracking. The experimental results had conducted on the OTB-2013 dataset. The two-layer convolutional feature and the HOG feature are adaptively fused, so that the more discriminative single feature fusion weight is greater, which better expresses the appearance model of the object, and shows strong object tracking accuracy in scenes such as complex background, the disappearance of the object, light change, fast movement, and rotation.