Video saliency object detection (VSOD) algorithms in the Internet of Things (IoT) frequently encounter challenges related to both detection accuracy and processing speed. Existing approaches have predominantly focused on enhancing detection accuracy, often resulting in larger models that compromise detection speed. In response, this paper introduces an efficient VSOD algorithm designed to achieve an optimal balance between detection accuracy and processing speed. The proposed algorithm comprises two primary modules: the Inter-frame Self-Attention (IFSA) module and the Attention ConvGRU (AttConvGRU) module. The IFSA module operates at deeper network layers, leveraging inter-frame interactions to accurately localize salient objects and predict their motion trends. Meanwhile, the AttConvGRU module is utilized to extract temporal features from shallow layers by incorporating an attention mechanism between ConvGRU units, thereby enhancing the temporal feature extraction capabilities. Experimental evaluations on five benchmark datasets demonstrate that the proposed algorithm achieves state-of-the-art detection accuracy and operates at a speed of 57 frames per second, effectively meeting real-time requirements for video salient object detection.