Master Assignment
[M][B] Artificial intelligence: Computer vision for image and video understanding
Type: Master EE/CS/HMI
Period: TBD
Student: (Unassigned)
If you are interested please contact :
Description:
Example of projects that I work on;
Human behaviour recognition
Human action and behavior recognition has a wide range of real-world applications: health monitoring, human-computer interaction, to name a few. Video-based human behavior recognition is one of the most complex tasks in computer vision. It usually involves the detection and classification of spatio-temporal behavioral patterns. Egocentric vision is a field that focuses on developing frameworks for analyzing and understanding human behavior from data collected from a first-person view, i.e., collected by wearable cameras.
Related works:
- Talavera, E., Wuerich, C., Petkov, N., & Radeva, P. (2020). Topic modelling for routine discovery from egocentric photo-streams. Pattern Recognition, 104, 107330.
- Cartas, Alejandro, et al. "Understanding event boundaries for egocentric activity recognition from photo-streams." International Conference on Pattern Recognition. Springer, Cham, 2021.
Human related crime recognition
Automatic detection of anomalies captured by surveillance environments is essential to streamline the otherwise laborious approach. To date, UCF-Crime is the largest dataset available for automatic visual anomaly analysis and consists of real-world crime scenes of various categories. Recently, HR-Crime has been introduced as a subset of the UCF-Crime dataset suitable for human-related anomaly detection tasks. Previous work in this field relied on descriptors such as skeleton trajectories, video depth, audio signals and radar for recognition of different human activities[1]. I find it interesting in automatic crime recognition and understanding the surrounding context.
Related works:
- Boekhoudt, Kayleigh, et al. "HR-Crime: Human-Related Anomaly Detection in Surveillance Videos." International Conference on Computer Analysis of Images and Patterns. Springer, Cham, 2021.
- Matei, Alina-Daniela, Estefania Talavera, and Maya Aghaei. "Crime scene classification from skeletal trajectory analysis in surveillance settings." arXiv preprint arXiv:2207.01687 (2022).
Multimodal analysis for scene recognition
Indoor scene recognition is a growing field with great potential for behavioural understanding, robot localization, and elderly surveillance, among others. I am interested in combining data modalities for scene identification.
Related works:
- Glavan, Andreea, and Estefanía Talavera. "InstaIndoor and multi-modal deep learning for indoor scene recognition." Neural Computing and Applications 34.9 (2022): 6861-6877
Other topics: medica image analysis, event-based cameras for video analysis, sentiment analysis from visuals and text, among others.