In realtime multi-object tracking (MOT), modeling the trajectory of objects over a video clip is often a challenging task. Currently, one popular approach is the tracking-by-detection (TBD) paradigm, mainly relying on matching detection boxes with the tracklets. However, detection boxes generally do not include heavily occluded targets, as they could be filtered out during the non-maximum suppression (NMS) stage. This leads to the ineffective tracking of occluded targets. In fact, these heavily occluded targets could be detected within the proposal regions before the NMS. Motivated by this, we propose a Robust Enhancement technique for TBD-based multi-object Tracking methods by associating proposal regions with tracklets, named RETrack. For tracklets that cannot be matched with detection boxes, RETrack associates them with the proposal regions, thereby recovering the targets. We believe RETrack can enhance the performance of multi-object tracking tasks while maintaining real-time capability. When applied to 4 different real-time trackers, RETrack achieves consistent improvement on the MOTA, IDF1 and HOTA scores. We integrate RETrack with ByteTrack and apply it on the popular MOT challenge benchmark and achieve 78.7 MOTA, 76.8 IDF1, and 62.4 HOTA on the MOT17 dataset and 74.8 MOTA, 74.9 IDF1, and 60.8 HOTA on the MOT20 dataset, respectively.