The radar echoes from weak pedestrians are easily submerged by strong background clutters in urban road traffic scenarios, resulting in the failure of trajectory tracking and prediction. To address this problem, this article proposes an integrated trajectory tracking and prediction method for weak pedestrians based on frequency-modulated continuous wave-multiple input multiple output millimeter wave radar. Firstly, the recursive Bayesian track-before-detect algorithm is utilized to directly extract target motion trajectory from the non-thresholding three-dimensional radar raw spectrum, which avoids the tracking performance degradation caused by the information loss from the traditional threshold-decision process. On this basis, this article proposes a Transformer-based end-to-end trajectory prediction model to further explore the spatiotemporal correlations of tracking trajectory and achieves accurate trajectory prediction of a weak pedestrian. Experimental results show that, when the signal-to-noise ratio is greater than - 20 dB, the average displacement error and final displacement error of the predicted trajectory of the proposed method are less than 0. 706 and 1. 215 m, respectively, which are all superior to traditional methods such as Gaussian process and long short-term memory network.