Hi Tiziana,
The “vision.ForegroundDetector” algorithm requires a few initial frames containing only the background in order to learn and create a background model. These frames are used to estimate the background and differentiate it from the moving objects (cells) in subsequent frames.
In your case, where the cells you want to detect are always moving and there are no frames with only the background, you can still use the “vision.ForegroundDetector” algorithm, but you might need to adjust the parameters and experiment with different settings to achieve satisfactory results.
Here are a few suggestions to handle the scenario where the cells are always moving:
1. Adjust the number of training frames (“NumTrainingFrames”): Reduce the number of training frames to a smaller value, especially if you have a limited number of frames available without the cells. But if it is not a short video, you can increase the number of training frames, the algorithm will have more samples of the background to learn from.
2. Experiment with different parameter values: You can try adjusting other parameters of the “vision.ForegroundDetector” algorithm, such as “NumGaussains”, “LearningRate” and “MinimumBackgroundRatio”. Experimenting with different values can help you find the optimal settings for your specific video. In this case, since the background is static, you can reduce the value of “NumGaussians”. To understand more about choosing the hyperparameters for the detector, you can refer to the following MATLAB Answer:
Hope this helps!
