Feature-based Particle Filter for Multiple Objects Tracking

Abstract

This paper proposes an advanced particle filter for multi-target tracking with speed-up robust features. In this study, a mixture of the Gaussian Background Model and the SURF algorithm are used for target representation and localization. This approach transforms an image into a large collection of local feature vectors, each of which is invariant to the images translation, scaling, and rotation. Additionally, it is also partially invariant to illumination changes and affine or 3D projection. Lastly, NN algorithm is used for segmenting multiple objects into a single-object state space. Several experimental results show that the proposed algorithm has good performance for object tracking in the presence of object translation, rotation and partial occlusion. Overall, this approach makes the system robust to occlusions and allows false positive detections in the background to be identified and removed.