Silhouette Orientation Volumes for Efficient Fall Detection in Depth Videos

Abstract

A novel method to detect human falls in depth videos is presented in this paper. A fast and robust shape sequence descriptor, namely the Silhouette Orientation Volume (SOV), is used to represent actions and classify falls. The SOV descriptor provides high classification accuracy even with a combination of simple associated models, such as Bag-of-Words and the Naïve Bayes classifier. Experiments on the public SDU-Fall dataset show that this new approach achieves up to 91.89% fall detection accuracy with a single-view depth camera. The classification rate is about 5% higher than the results reported in the literature. An overall accuracy of 89.63% was obtained for the six-class action recognition, which is about 25% higher than the state of the art. Moreover, a perfect silhouette-based action recognition rate of 100% is achieved on the Weizmann action dataset.

Fall detection without people: A simulation approach tackling video data scarcity

“…Table 3 summarises the results of the proposed method when tested against two public datasets alongside the performance of other methods tested on the same datasets. The proposed methodology outperforms previous works on both datasets in terms of accuracy, precision and 520 specificity and its sensitivity is similar to all but (Akagündüz et al, 2017) where more falls are detected, but more ADLs are also detected as falls. These results show that the simulated approach has almost equal or better performance when tested against these public datasets.…”

Radar and RGB-Depth Sensors for Fall Detection: A Review

“…Two other works have used this dataset after Ma et al [88], who originally collected the data. In particular, the accuracy originally obtained on this dataset was around 87%, and it has been improved to 89% in [90] and almost 92% in [92]. Regarding radar sensors, to authors' best knowledge, openly accessible datasets of radar signatures for fall vs non-fall detection are at the moment not available to the research community.…”

“…Considering a dataset constituted by 8 activities, the system achieves a sensitivity of 90% and a specificity of 100% for the falling events. A shape sequence descriptor, namely the Silhouette Orientation Volume (SOV), has been proposed in [92]. This descriptor has been associated to BoW models, for which the codebook has been built considering k-medoids clustering technique, and then Naïve Bayes classifier can be used to recognize fall related actions of the SDUFall dataset and also more general actions of the Weizmann dataset [93].…”

Queueing Detection System Using an Integrated Pedestrian Flow and Optical Regression

“…As shown in Figure 1. This silhouette would be utilized to approve the blobbing object is definitely a human [10], or to identify the silhouette of the bounding box from where additional individual's features can be obtained for that goal [11]. Because some of morphological procedures do not assure that every individual translate to accurately one segment (or blob), an additional pass should be accomplished where segments which are sufficiently adjacent have been combined together.…”