Human Movement Detection
In this post, we talk about Human Movement Detection.
The use of the low extrema point, which is the person’s foot location, to detect and compute the real distance in the 3D space caused the poor segmentation result. used the background subtraction method based on the Mixture of Gaussian 2 (MOG2). The algorithm is highly reliable to create the background. However, the segmentation method is affected critically by the light condition and the contouring result. The poorly done segmentation leads a high error in the distance approximation.
Therefore, use the alternative methods to capture better low extrema point in the image. keep the original Mixture of Gaussian (MOG) background subtraction and add the pixel location detection of the ROI to adjust the outcome of the false detection. The major false detection comes from the poor segmentation as shown in Figure and the division of the multi ROI area. The algorithm only considers the maximum area of the intensity change for the mass center detection. Therefore, the lower extrema point is only local lower extrema point and the real lowest extrema point is undetected as shown in Figure(b). By using manual pixel location detection of the entire ROI, we can find the real extrema point of the ROI and reduce the false detection of the foot location. For the true detection of the lowest extrema point as shown in Figure , we used pixel by pixel subtraction after thresholding. The result is much better, but computationally the process is expensive.
