The challenge of joint detection and tracking of multiple extended targets (ETs) arises in many radar applications, especially for extended stealth targets (ESTs) with small signal-to-noise ratio (SNR). Recently, the multi-Bernoulli (MB) filter-based random matrix model (RMM) has been proposed for tracking ellipsoidal ETs. The MB-RMM filter depends on the measurements (position and range rate), which are extracted by thresholding the received signal. It is acceptable if SNR is high. The threshold must be less than the sufficient probability to avoid a high rate of false detections. This assumption is unsuitable for tracking extended-stealth targets (ESTs) with unknown detection probability. To this point, we address a track-before-detect-MB-RMM approach, which is an efficient way to solve the unknown detection profile issue. There is a considerable advantage of using unthresholded data in simultaneous detection and tracking. In this scenario, although the extension ellipsoid is efficient, it may not be accurate enough because of a lack of useful information, such as size, shape, and orientation. Therefore, we introduce a filter composed of subellipses where each one is represented by an RMM. Furthermore, a sequential Monte Carlo implementation is applied to estimate nonlinear kinematic ESTs’ state. The results confirm the effectiveness and robustness of the proposed filter.