Hands-On Artificial Intelligence for IoT

HAR from videos

Another way to detect human activity is through videos. In this case, we will have to use a DL model such as CNN to get good results. A good dataset for classified videos is available from Ivan Laptev and Barbara Caputo (http://www.nada.kth.se/cvap/actions/). It contains six types of action: walking, jogging, running, boxing, hand waving, and hand clapping, in different scenarios. Each video has been recorded using a camera with 25 fps. The spatial resolution is 160 × 120, and of an average length of four seconds. It has in total 599 videos with about 100 for each of the six categories.

One of the problems with video data is that it is computationally expensive, thus it will be important to reduce the dataset, and a few ways ...