Multi-oriented run length based static and dynamic features fused with Choquet fuzzy integral for human fall detection in videos

Abstract

Since a huge part of elderly people are living alone, assisted-living tools have become an essential in-home telemonitoring device. Hence, this paper proposes an automatic human fall detection in videos. In order to improve the system reliability, a new shape descriptor called multi-oriented run length (MORL) is proposed. This descriptor is exploited in a proposed scheme to generate static and dynamic features to represent human falls with complementary information. The generated static and dynamic features are fused through the Choquet fuzzy integral. Experimental results conducted on three well-known datasets containing almost 1300 video segments show an interesting adaptation of the proposed approaches. More precisely, the proposed MORL descriptor shows its superiority against known descriptors such as LBP and HOG. Moreover, Choquet fuzzy integral significantly improves the results versus standard combiners. In general, the obtained results highlight the reliability of the proposed system versus recent studies for human fall detection.