Skip to main content

Table 1 Comparison of object detection methods in terms of accuracy and computational time

From: Human detection in surveillance videos and its applications - a review

Methods Accuracy Computational time Comments
Background subtraction Mixture of Gaussian model [210] Moderate Moderate Simple implementation and good performance but not so well with dynamic background. It requires parameters to be defined by the practitioners. It can capture multi-modal scenarios
Non-parametric background model [1217] Moderate to high Low to moderate In dynamic background scenarios, NP performs very well compared to MoG-based algorithm. It requires significant post-processing. In occlusion situation, it does not perform well compare to MoG
Temporal differencing [19, 20, 47, 69] High Low to moderate Very good with sudden illumination changes in indoor environment
Warping background [21] High Moderate to high Good in outdoor environment with high background motion. It does not handle occlusion well. Some variations are computationally intensive
Hierarchical background model [22, 6668] High Low to moderate Make use of both block-based and pixel-based approaches. May be quicker than pixel-based approach, but quality could be compromised
Optical flow [18, 2326] Moderate High Good with camera motion and crowd detection but highly computation intensive
Spatio-Temporal filter [2737, 70] Moderate to high Low to moderate Works well for low-resolution scenarios but suffers from noise issues