Compressive spectral imaging (CSI) systems capture three dimensional (3D) spatio-spectral information of a scene using two-dimensional (2D) coded focal plane array measurements. Some of these architectures have been adapted to effectively capture temporal sequences of dynamic spatial-spectral sources, at high frame rates, which has enabled motion detection in applications such as surveillance. However, the drawback of state-of-the-art compressive motion detection techniques over spectral video is that they first require the reconstruction of each frame. So that, the computational complexity grows with the number of frames. Hence, this work proposes a method to detect the salient motion regions directly on the compressive domain to avoid the reconstruction step. Specifically, the proposed approach uses a recursive instantaneous model that analyzes motion within consecutive CS projections. Simulation results show that the proposed approach reduces the motion detection time in up to 99% with the same precision compared with methodologies that work on the reconstructed data.