Integrating sensing capabilities into base stations (BSs) represents a crucial research direction for 6G wireless communications. Given limited transmit power and the complex perception environment, it is necessary to enhance the sensing performance of BSs in low signal-to-noise ratio (SNR) conditions. Aiming at this challenge, this paper proposes a track-before-detect (TBD) scheme based on the extended particle filter (EPF) and multi-BS collaboration. In this scheme, the likelihood ratio derived from the incoherent fusion of multi-BS signals is utilized in the EPF to jointly perform target number estimation and trajectory tracking. In the EPF, to improve tracking accuracy and accelerate the detection of newborn targets, sampling approaches that leverages the characteristics of OFDM signals are introduced to generate independent measurements. Then, an efficient importance density is designed for managing target birth, survival, and death respectively to enhance the effectiveness of particles. Finally, the particle parameter sharing (PPS) strategy is proposed to reduce computational complexity, addressing key challenges in PF algorithms. With the well-designed measurements and importance density, the proposed EPF-TBD exhibits excellent detection and tracking performance under low SNR conditions, outperforming existing DBT and TBD algorithms in simulations.