Iterative scene learning in visually guided persons' falls detection

This article describes a fast real time computer vision algorithm able to detect humans' falls in complex dynamically changing visual conditions. The algorithm exploits single cameras of low cost while it requires minimal computational cost and memory requirements. Due to its affordability it can be straightforwardly implemented in large scale clinical institutes/home environments. In this paper, we evaluate the performance of this algorithm into two different real-world conditions. The evaluation was performed for long time and concerns robustness compared to other humans' activities, false positive/negative estimates, all in real time.