This paper presents a technique to improve Gaussian mixture models for robust object detection by modifying the new model induction logic and using intensity difference thresholding to detect objects from one or more background models. The proposed method eliminates drawbacks of poor Gaussian mixture quality, susceptibility to background/foreground data proportion, and instability with varying operating environments. Quantitative and qualitative evaluations on test video sequences show the proposed technique achieves lower error rates and better visual results compared to existing methods.