Description

Detecting human instances in static and/or dynamic scenes is crucial for developing solutions for open problems such as video understanding, anomaly detection, compression, and restoration. Segmentation is necessary to retrieve representative instances of human beings, resulting in the extraction of regions of interest used for solutions focusing solely on the semantic behavior of humans in their environment. Annotation complements segmentation by matching the regions of interest obtained from each image during segmentation. The major challenge is correctly matching instances considering the challenges that may arise when collecting the image sequence.

The objective of this study is to implement an intelligent model for segmenting regions representing human beings in image sequences and connect the model with a bio-inspired module for region matching.

References:

[1] I. Katircioglu, H. Rhodin, J. Sporri, M. Salzmann, and P. Fua, “Human Detection and Segmentation via Multi-view Consensus,” 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Oct. 2021, doi: 10.1109/iccv48922.2021.00285

[2] M. H. Bhatti, M. Azeem, and H. Younis, “Object Segmentation in Video Sequences by using Single Frame Processing,” 2019 13th International Conference on Open Source Systems and Technologies (ICOSST), Dec. 2019, doi: 10.1109/icosst48232.2019.9043975