Improved Resampling Algorithms Based on Deterministic and Partial Resampling

Abstract

Particle Filter is widely used to track objects in autonomous driving. Resampling is one of the most important steps in the particle filter. In implementations, resampling is regarded as a bottleneck because of the increased complexity. To handle such kind of difficulties, an improved resampling algorithm is proposed. Firstly, the most frequently applied resampling algorithms for particle filters are introduced in this paper. A theoretical analysis is provided and the differences among these resampling algorithms are revealed. This facilitates a comparison of the algorithms about their resampling quality and computational complexity. Through the simulations, the results in the theoretical analysis are verified. It is found that optimized partial deterministic resampling is suitable for tracking in terms of a large amount of objects.