Target detection and tracking in real-time video streaming is one of the core tasks in the field of computer vision, and its performance directly affects the effectiveness of intelligent surveillance, automatic driving and other applications. In this paper, a three-frame frame difference method incorporating Sobel operator is proposed to optimize the boundary extraction by combining morphological filtering. Meanwhile, the Mean Shift-based target tracking algorithm is improved by introducing a recognizable interferer model with a dual attention mechanism to enhance the tracking robustness. Experiments show that the improved inter-frame difference method achieves F-measure of 93.38%, 96.20% and 91.83% on Highway, Road and Highway2 datasets, respectively, and the detection frame rate is improved to 32 f/s. The improved Mean Shift algorithm improves the tracking accuracy by 2.67%-3.09% and the success rate by 0.98%-1.75% over the mainstream algorithm on the UAV123 and VOT2024 datasets. In addition, the multi-target tracking experiments achieve MOTA of 75.76% and 80.61% on the MOT16 and MOT17 datasets, respectively, validating the comprehensive performance advantages of the algorithms. This study provides an efficient and robust solution for target detection and tracking in real-time video streaming.