Fallen Person Detection for Mobile Robots Using 3D Depth Data

Abstract

Falling down and not managing to get up again is one of the main concerns of elderly people living alone in their home. Robotic assistance for the elderly promises to have a great potential of detecting these critical situations and calling for help. This paper presents a feature-based method to detect fallen people on the ground by a mobile robot equipped with a Kinect sensor. Point clouds are segmented, layered and classified to detect fallen people, even under occlusions by parts of their body or furniture. Different features, originally from pedestrian and object detection in depth data, and different classifiers are evaluated. Evaluation was done using data of 12 people lying on the floor. Negative samples were collected from objects similar to persons, two tall dogs, and five real apartments of elderly people. The best feature-classifier combination is selected to built a robust system to detect fallen people.

Fast and robust detection of fallen people from a mobile robot

“…In [22], the authors propose a pipeline working on just single RGB images extending a deformable part-based model to the multi-view case for viewpoint invariant lying posture detection. Like us, [23] proposes a pipeline working on single depth images. Putative candidates are found by means of a segmentation phase based on an Euclidean clustering.…”

“…On the contrary, this work specifically addresses this problem by concatenating two classifiers. Unfortunately, neither the code or dataset of [23] are available making a direct comparison impossible. Finally, in [24], a method for detecting and locating the head of a person lying on the floor by means of a RGB-D sensor is proposed.…”

“…In particular, we present the results for each of the 3 contributions, the single-view detector and the two modules of the multi-view analyser: the map validation and the detection merging from multiple vantage points. Furthermore, given the impossibility to compare directly with [23], the comparison baseline (B) is a simple approach not exploiting patches. It finds putative clusters by means of the Euclidean cluster extraction with a distance threshold of 0.10 m, which is really low considering a voxel resolution of 0.06 m and far less than the one required by our approach (1 m).…”

Automated Fall Detection Technology in Inpatient Geriatric Psychiatry: Nurses’ Perceptions and Lessons Learned

“…It has been suggested that using multiple Kinect sensors may be necessary to make further improvements in these areas (Stone & Skubic, 2015), but interference between the multiple near-IR light sources may present a challenge for this approach. One research team has mounted the Kinect on a robot that patrols its environment searching for fallen people (Volkhardt, Schneemann, & Gross, 2013), but this approach may not be tolerated well by older adults, particularly those experiencing dementia, paranoia, or delusions.…”

Online pedestrian group walking event detection using spectral analysis of motion similarity graph

“…Detection of events has gained lots of attention in video surveillance domain, it is the case of research related to the identification of lying pose recognition [15,13], detection of people running [4,19], crowd safety [7,16], etc. In videos that include civil interactions, modeling the social behaviors of people plays an important role in describing the individual and group behaviors on crowded scenes [18].…”