Abstract: Motion object detection is a critical research area in computer vision, aiming to identify moving objects from videos or continuous image sequences. However, traditional motion object detection algorithms often suffer from low detection accuracy in complex environments, such as dynamic backgrounds and varying illumination conditions. To address this limitation, this paper proposes a Gaussian Mixture Model-based Background Subtraction method with Temporal and Spatial Attention mechanisms (TSA-BSGMM). Built upon the traditional Gaussian mixture model-based background subtraction framework, the proposed method incorporates a spatial attention mechanism in the background modeling phase to enhance model precision. During the background-foreground segmentation stage, morphological processing is applied to suppress noise and reinforce object shapes. Furthermore, a multi-level temporal attention mechanism is introduced in the background update phase, enabling the model to rapidly adapt to abrupt environmental changes, such as sudden illumination variations. Experimental results show that the TSA-BSGMM achieves an accuracy of 87.92% on a self-built dataset. Moreover, within 13 frames after a sudden change in illumination, the detection accuracy recovers to over 90% of the pre-change level, demonstrating significant superiority over existing state-of-the-art motion object detection methods.