Three Ways to Improve the Performance of Real-Life Camera-Based Fall Detection Systems

Figure 4

Overview of the contents of the feature vector (FV) used by the support vector machine (SVM; Section 3.3). The complete video is split into discrete, one-second time slots. One FV is created for each time slot. A FV contains information about the current time slot and (combinations of) other time slots. Each FV part consists of 10 features as shown (AR = aspect ratio, CAR = change of AR, FA = fall angle, CS = center speed, and HS = head speed) [28].