• artificial intelligence
• computer vision
• deep learning
• fallen people detection

This thesis work is about recognizing fallen people through a Computer Vision approach. Falling is one of the most common causes of injury in all ages, especially in the elderly, where it is most frequent and severe. For this reason, a method that can detect a fall in real-time can be helpful to avoid more severe damage. Some methods use sensors, wearable devices, or video cameras with particular features such as infrared or depth cameras. However, in this work, we used a vision-based approach to exploit classic video cameras that are more accessible and widespread. A limitation of this method is the lack of generalization to unseen environments. This is due to the error generated during the object detection and, overall, for the unavailability of a large-scale dataset specialized in fallen detection problems with different environments and fallen types. In this work, we try to solve these problems by an object detector trained using a virtual-world dataset in addition to real-world images. We tested our models, and we noted that using synthetic images improves generalization from these results.