Vision- and Sensor-based Activity Analysis: Future Scopes
Gait or walking pattern is an important biometric cue for various applications. Vision- based gait recognition has the potential to recognize subject even with a low-resolution image sequence and it can be captured without the subject’s cooperation at a distance. Gait- based biometrics at a distance have applications in surveillance, healthcare, medical applications (for example, rehabilitation process), forensic and even criminal investigation. However, there are some challenges in gait recognition to solve like clothing issues (e.g., jacket, long skirt, very loose-outfit), view-invariance issues, gait with carried objects (e.g., bags, umbrella, backpack), and occlusions. Among them, occlusion issue is a less-explored challenge even though it is a genuine problem in real-life situations. Usually, several body-parts of subjects are often occluded due to the beams, pillars, car, and tree or with another walking person. Therefore, it is difficult to extract an unoccluded sequence for a full gait cycle, which is required to gait recognition for most of the current approaches. In this speech, a novel approach based on a conditional deep generative adversarial network is presented . The proposed generative network exploits an adversarial loss that is based on triplet hinge loss along with Wasserstein GAN. In this method, we reconstruct the silhouette sequence from an occluded sequence, and recognize gait even in the presence of different types of occlusions. Note that wearable sensor-based gait methods have been explored as well  though occlusions are not a concern similar to the video-based gait approaches.
MZ Uddin, D Muramatsu, N Takemura, MAR Ahad, Y Yagi, “Spatio-temporal silhouette sequence reconstruction for gait recognition against occlusion,” IPSJ Transactions on Computer Vision and Applications, Springer, 11(1), 2019.
TT Ngo, MAR Ahad, AD Antar, M Ahmed, D Muramatsu, Y Makihara, Y Yagi, S Inoue, T Hossain, and Y Hattori, “OU-ISIR Wearable Sensor-based Gait Challenge: Age and Gender”, 12th IAPR Intl. Conf. on Biometrics (ICB), IEEE, 2019.
Md Atiqur Rahman Ahad, SMIEEE; Professor, University of Dhaka (DU); Specially Appointed Associate Professor, Osaka University. He did B.Sc.(Honors) & Masters (DU), Masters (University of New South Wales), PhD (Kyushu Institute of Technology), JSPS Postdoctoral Fellow and Visiting Researcher. His authored books are: “Motion History Images for Action Recognition and Understanding”, in Springer; “Computer Vision and Action Recognition”, in Springer; “IoT-sensor based Activity Recognition”, in Springer (in press). He has been authoring/editing a few more books. He published 130+ peer-reviewed papers, 60+ keynote/invited talks, 25+ Awards/Recognitions. He is Editorial Board Member of Scientific Reports, Nature; Assoc. Editor of Frontiers in Computer Science; Editor of Int. Journal of Affective Engineering; Encyclopedia of Computer Graphics and Games, Springer; Editor-in- Chief: Int. Journal of Computer Vision & Signal Processing http://cennser.org/IJCVSP; General Chair: 9th ICIEV http://cennser.org/ICIEV; 4th IVPR http://cennser.org/IVPR; 2 nd ABC https://abc-research.github.io, Guest-Editor: Pattern Recognition Letters, Elsevier; JMUI, Springer; JHE, Hindawi; IJICIC; Member: OSA, ACM, IAPR. More: http://AhadVisionLab.com