Person Re-Identification Using Deep Modeling of Temporally Correlated Inertial Motion Patterns

Bodyprint-A Meta-Feature Based LSTM Hashing Model for Person Re-Identification

Person re-identification is concerned with matching people across disjointed camera views at different places and different time instants. This task results of great interest in computer vision, especially in video surveillance applications where the re-identification and tracking of persons are required on uncontrolled crowded spaces and after long time periods. The latter aspects are responsible for most of the current unsolved problems of person re-identification, in fact, the presence of many people in a location as well as the passing of hours or days give arise to important visual appearance changes of people, for example, clothes, lighting, and occlusions; thus making person re-identification a very hard task. In this paper, for the first time in the state-of-the-art, a meta-feature based Long Short-Term Memory (LSTM) hashing model for person re-identification is presented. Starting from 2D skeletons extracted from RGB video streams, the proposed method computes a set of novel meta-features based on movement, gait, and bone proportions. These features are analysed by a network composed of a single LSTM layer and two dense layers. The first layer is used to create a pattern of the person’s identity, then, the seconds are used to generate a bodyprint hash through binary coding. The effectiveness of the proposed method is tested on three challenging datasets, that is, iLIDS-VID, PRID 2011, and MARS. In particular, the reported results show that the proposed method, which is not based on visual appearance of people, is fully competitive with respect to other methods based on visual features. In addition, thanks to its skeleton model abstraction, the method results to be a concrete contribute to address open problems, such as long-term re-identification and severe illumination changes, which tend to heavily influence the visual appearance of persons.

An Acceleration Based Fusion of Multiple Spatiotemporal Networks for Gait Phase Detection

Human-gait-phase-recognition is an important technology in the field of exoskeleton robot control and medical rehabilitation. Inertial sensors with accelerometers and gyroscopes are easy to wear, inexpensive and have great potential for analyzing gait dynamics. However, current deep-learning methods extract spatial and temporal features in isolation-while ignoring the inherent correlation in high-dimensional spaces-which limits the accuracy of a single model. This paper proposes an effective hybrid deep-learning framework based on the fusion of multiple spatiotemporal networks (FMS-Net), which is used to detect asynchronous phases from IMU signals. More specifically, it first uses a gait-information acquisition system to collect IMU sensor data fixed on the lower leg. Through data preprocessing, the framework constructs a spatial feature extractor with CNN module and a temporal feature extractor, combined with LSTM module. Finally, a skip-connection structure and the two-layer fully connected layer fusion module are used to achieve the final gait recognition. Experimental results show that this method has better identification accuracy than other comparative methods with the macro-F1 reaching 96.7%.

Image-Based Somatotype as a Biometric Trait for Non-Collaborative Person Recognition at a Distance and On-The-Move

It has recently been shown in Re-Identification (Re-ID) work that full-body images of people reveal their somatotype, even after change in apparel. A significant advantage of this biometric trait is that it can easily be captured, even at a distance, as a full-body image of a person, taken by a standard 2D camera. In this work, full-body image-based somatotype is investigated as a novel soft biometric feature for person recognition at a distance and on-the-move. The two common scenarios of (i) identification and (ii) verification are both studied and evaluated. To this end, two different deep networks have been recruited, one for the identification and one for the verification scenario. Experiments have been conducted on popular, publicly available datasets and the results indicate that somatotype can indeed be a valuable biometric trait for identity recognition at a distance and on-the-move (and hence also suitable for non-collaborative individuals) due to the ease of obtaining the required images. This soft biometric trait can be especially useful under a wider biometric fusion scheme.

Person Re-ID by Fusion of Video Silhouettes and Wearable Signals for Home Monitoring Applications

The use of visual sensors for monitoring people in their living environments is critical in processing more accurate health measurements, but their use is undermined by the issue of privacy. Silhouettes, generated from RGB video, can help towards alleviating the issue of privacy to some considerable degree. However, the use of silhouettes would make it rather complex to discriminate between different subjects, preventing a subject-tailored analysis of the data within a free-living, multi-occupancy home. This limitation can be overcome with a strategic fusion of sensors that involves wearable accelerometer devices, which can be used in conjunction with the silhouette video data, to match video clips to a specific patient being monitored. The proposed method simultaneously solves the problem of Person ReID using silhouettes and enables home monitoring systems to employ sensor fusion techniques for data analysis. We develop a multimodal deep-learning detection framework that maps short video clips and accelerations into a latent space where the Euclidean distance can be measured to match video and acceleration streams. We train our method on the SPHERE Calorie Dataset, for which we show an average area under the ROC curve of 76.3% and an assignment accuracy of 77.4%. In addition, we propose a novel triplet loss for which we demonstrate improving performances and convergence speed.

Characterizing Human Box-Lifting Behavior Using Wearable Inertial Motion Sensors

Although several studies have used wearable sensors to analyze human lifting, this has generally only been done in a limited manner. In this proof-of-concept study, we investigate multiple aspects of offline lift characterization using wearable inertial measurement sensors: detecting the start and end of the lift and classifying the vertical movement of the object, the posture used, the weight of the object, and the asymmetry involved. In addition, the lift duration, horizontal distance from the lifter to the object, the vertical displacement of the object, and the asymmetric angle are computed as lift parameters. Twenty-four healthy participants performed two repetitions of 30 different main lifts each while wearing a commercial inertial measurement system. The data from these trials were used to develop, train, and evaluate the lift characterization algorithms presented. The lift detection algorithm had a start time error of 0.10 s ± 0.21 s and an end time error of 0.36 s ± 0.27 s across all 1489 lift trials with no missed lifts. For posture, asymmetry, vertical movement, and weight, our classifiers achieved accuracies of 96.8%, 98.3%, 97.3%, and 64.2%, respectively, for automatically detected lifts. The vertical height and displacement estimates were, on average, within 25 cm of the reference values. The horizontal distances measured for some lifts were quite different than expected (up to 14.5 cm), but were very consistent. Estimated asymmetry angles were similarly precise. In the future, these proof-of-concept offline algorithms can be expanded and improved to work in real-time. This would enable their use in applications such as real-time health monitoring and feedback for assistive devices.