Person Tracking
Project Details
Video surveillance is intended to protect people and property from harm.Multiple Camera Re-Identification
Person re-identification is the task of associating images of the same person taken from different cameras or from the same camera in different occasions.
- Types of Algorithms : Convolutional Neural Networks
- Technologies : Python, Triton, DALI
- Modules : Object Detection, Object Tracking, Object Re-Identification
- SOTA Networks : Transformers
- Association Challenge : Hungarian Algorithm
Tracking allows us to receive all the trajectories of movement of anyone in the scene and identify one person from another. From one second to another second, but the challenges start to appear when we have multiple cameras.
The image retrieval task aims to find the most similar object to the query image. Despite active development of retrieval models it still remains a challenging task mainly due to large intra-class variance caused by changes in view angle, lighting, background clutter or occlusion, while inter-class variance may be relatively low.
Tracking a person of interest or a group of people requires to track objects in space and time. The goal is to follow the trajectories of people, pedestrians, while keeping their identities discriminated as they are moving throughout a video sequence.
Increased efficiency for security by automatization. Crowd psychology. Behavior prediction. Real-time preventative action.