An Empirical Study Towards Understanding How Deep Convolutional Nets Recognize Falls

Abstract

Detecting unintended falls is essential for ambient intelligence and healthcare of elderly people living alone. In recent years, deep convolutional nets are widely used in human action analysis, based on which a number of fall detection methods have been proposed. Despite their highly effective performances, the behaviors of how the convolutional nets recognize falls are still not clear. In this paper, instead of proposing a novel approach, we perform a systematical empirical study, attempting to investigate the underlying fall recognition process. We propose four tasks to investigate, which involve five types of input modalities, seven net instances and different training samples. The obtained quantitative and qualitative results reveal the patterns that the nets tend to learn, and several factors that can heavily influence the performances on fall recognition. We expect that our conclusions are favorable to proposing better deep learning solutions to fall detection systems.

Skeleton-based Privacy-Preserving Smart Activity Sensor for Senior Care and Patient Monitoring

“…Table 1 shows the results of our fall detection algorithm and results reported in [2]. Our method achieves higher frame-level results in the 2-fold test and slightly lower than theirs for the 4-fold test.…”

“…1) Following [2], 96 video clips are extracted with 45 frames before the start of the fall action and 15 frames after the end of fall action. In [2], frame-level accuracy is used to evaluate the performance of fall detection, i.e., each frame is labelled as fall or normal frame. The training and testing split also has two setups, 2-fold cross-validation and 4-fold cross-validation.…”

“…We also test the performance of our event-level fall detection algorithm, which achieves 100% accuracy for both 2-fold and 4-fold tests. No event-level result was reported in [2]. 2) Following [3], 127 videos are randomly selected as training and testing set in a 2-fold cross-validation setup.…”

Human Fall Detection Based on Re-Parameterization and Feature Enhancement

“…Taufeeque M et al [30] propose an LSTM-based approach to human pose estimation, which supports multi-camera and multi-target tracking. By studying potential fall recognition processes, Yan Z et al [31] reveal the patterns of network learning and the reasons that seriously affect the performance of model detection. Xiaoping Zhang et al [32] propose a new imagebased fall detection method in bus compartment scene, this method combines object detection and pose estimation, and introduce the fall identification network, and achieves 90% accuracy in the bus scene.…”