Adaptive Vector-Based Sample Consensus Model for Moving Target Detection in Infrared Video

The detection of moving targets in infrared video with competitive accuracy and less computation time is an intractable task for daily security. The background subtraction method is typically used for such tasks. However, owing to the particular characteristics of infrared videos, only a few techniques are suitable. Because most classic background models cannot deal with low signal-to-noise ratios and small targets, an adaptive vector-based background subtraction model is proposed to detect moving objects in infrared video. For each pixel, several filters are employed to take past values, and a vector is assigned to each filter to represent the information in the neighborhood of the pixel. Then, the series of vectors comprises the background model, and the collinearity between the vector of the current pixel and the vectors in the background model is calculated based on cosine similarity. The current pixel is classified as foreground or background according to the times of collinearity. Finally, a random update scheme is employed to update the model. Extensive qualitative and quantitative experimental results have revealed that the proposed technique can achieve competitive performance than existing unsupervised state-of-the-art algorithms for tackling low signal-to-noise ratio and small target.

Automated summary

This paper proposes an adaptive vector-based background subtraction model for detecting moving objects in infrared videos. By using multiple filters to capture past values and calculating collinearity between the current pixel and the background model based on cosine similarity, the proposed technique achieves competitive performance in detecting moving targets under low signal-to-noise ratio and small target conditions.