Modelling of the background (“uninteresting parts of the scene”), and of the foreground, play important roles in the tasks of visual detection and tracking of objects. This paper presents an effective and adaptive background modelling method for detectin

For the evaluation of performance against each image sequence, we use three terms (as in [27]): False Positive (FP), False Negative (FN), and total error for that image sequence (te). FP is the number of background pixels that are wrongly marked as foreground; FN is the number of foreground pixels that are wrongly marked as background; te is the sum of FP and FN for each image sequence. For the evaluation of overall performance, we use TE (the sum of total error for all seven image sequences). Because many methods do not work well for the LS sequence, we also use TE* (the sum of total error excluding the light switch image sequence) for the evaluation. For each result image, we eliminated isolated foreground pixels: those whose 4-connected foreground pixel number is less than 8. Methods ET

MO TOD LS WT C B FA TE TE* SACON f.pos.6087 4467 1 826 3989 327 2067 2066 2759 LP

Mixture of

Gaussian

Bayesian

decision

Eigen-

background te f. neg.pos.te te 12035 8046 pos.27053 112510 1028 15802 1664 3496 2091 2972 31422 156031065

0 895 25 986 1324 375 1322 3084 1999 2876 1898 2706 2935 6433 365 2390 2978 649 969 11478 1015617677 16353f. pos.neg.Wallflower f. pos.te

Table 1: Experimental results by different methods on Wallflower benchmarks.

20