Moving object detection with moving background using optic flow

Contour Detection of Multiple Moving Objects in Unconstrained Scenes using Optical Strain

“…One can easily reach a 100% recall value by setting all pixels to foreground, no matter the problem: of course, this solution is to be proscribed as it is not representative of the scene. This phenomenon happens with algorithm [13]: one can see from Fig. 1e that the method considers most of the pixels in the image as foreground, leading to a high Recall value (a perfect one in fact due to the BB computation).…”

“…In particular, we use the TV-L1 optical flow model provided in [19] because of its fast computation and good performance. For Patel and Parmar [13] and Tang et al [14] we use the parameter values proposed in the corresponding paper. For our method, we tried several values for the window size, from n = 2 to n = 20, and kept the one yielding to the best f 1 -score.…”

“…II: Precision, Recall, F 1 -score and Jaccard index of the obtained results from some sequences our dataset. We compare our model with the ones proposed in [13] and [14]. The last row shows the average over all the 76 sequences from the dataset.…”

“…We compare our model with the methods proposed by Patel and Parmar [13] and Tang et al [14], which also use the optical flow as input to obtain the moving objects of the scene. Both methods work on the spatial domain, taking into account only one frame each time.…”

“…1) Patel and Parmar [13]: This method proposes to compute the edges directly over the optical flow. In particular, they propose an optical flow thresholding method based on the mean and the standard deviation of each channel of the optical flow.…”