An attempt to introduce 4D tracing in detection of anomalous motion by clustering the temporal and spatial data and contouring 2D information belonging to the same timing frame

In this paper we suggest a new method for a fall detection, based on the spatiotemporal data clustering (4D tracking), collected during a fall incidence. The set of sequential frames has been extracted from the videos, then the key features have been extracted to classify falls and no-falls activities. Based on the study of the stability of the human body. The article proposes a new model of human posture representation of fall behaviour, taking into consideration the ‘inverted pendulum model’, and using multimedia analysis to observe and evaluate the continuity of the fall event in time series changes; the aim was to extract and construct the inverted pendulum structure of human posture in real-life scenarios. Firstly, a spatiotemporal evolution map of the change and variation of human posture is constructed through multimedia analysis Then 3D spatial information has been sliced into 2D. Contour plots are built from the spatial projection coordinates that belong to the same timing window. The timing window is determined based on the duration of the instability/transition phase. By contouring the spatial coordinates that belong to the same timing window we can determine whether the person is in a sitting or lying down position after their fall or whether the person is either moving or not during a prolonged period following the fall. Finally, computer vision when added to multimedia analytics, would further reduce the time consumption for 2D slicing of 3D information followed by the clustering of those 2D spatial projections (from different projections) with referent timing frames (coordinated extracted from time series extracted from video data).