The proposed algorithm was tested u

The proposed algorithm was tested using three video sequences. The images were acquired at a frame rate of 30 fps and have a spatial resolution of 480 × 640 pixels. Since the vessels move slowly in the image, we process every 5 frames, discarding the others. In addition, the images are subsampled with a factor 2 : 1 to reduce the computation time. The proposed algorithm performs in real time in a standard PC, despite the fact that it was programmed in Matlab. The following values were adopted in all the experiments: maximum vessel area, Amax = 6000 pixels, smallest distance between blobs, R = 10 pixels, time horizon, D = 5 subsampled frames, and threshold, D∗ = 4 subsampled frames. These values were chosen by trial and error and were not changed during the experiments. Figure 5 shows images extracted from three video sequences (left) and the output of the detection system (right),showing the ability of the system to cope with vessels of different sizes and large amounts of reflections and sky.
To assess the algorithm, the target position and size (bounding box) was manually annotated for all the test images. Then we compute recall and precision

ผลลัพธ์ (อังกฤษ) 3:

The proposed algorithm was tested using three video sequences. The images were acquired at a frame rate of 30 FPS and have. A spatial resolution of 480 × 640 pixels. Since the vessels move slowly in the image we process, every, 5 frames discarding. The others. In addition the images, are subsampled with a factor 2: 1 to reduce the computation time. The proposed algorithm. Performs in real time in a, standard PC despite the fact that it was programmed in Matlab. The following values were adopted. In all the experiments: maximum, vessel area Amax = 6000 pixels smallest distance, between, blobs R = 10 pixels time horizon,,, D = 5 subsampled, threshold frames and, ∗ D = 4 subsampled frames. These values were chosen by trial and error and were not. Changed during the experiments. Figure 5 shows images extracted from three video sequences (left) and the output of the. Detection system (right), showing the ability of the system to cope with vessels of different sizes and large amounts of. Reflections and sky.To assess the algorithm the target, position and size (bounding box) was manually annotated for all the test images. Then. We compute recall and precision.