Glosarium AI
Kamus lengkap Kecerdasan Buatan
Object Tracking
Algorithmic process of locating and tracking one or more moving objects in a video sequence frame by frame, using spatial and temporal information.
Kalman Filter
Recursive state estimation algorithm that predicts the future position of an object based on its past motion and corrects these predictions with new observations.
DeepSORT
Multi-object tracking algorithm that combines an object detector with a Kalman filter and a deep re-identification network to maintain object identities over long sequences.
Multi-Object Tracking (MOT)
Object tracking discipline that simultaneously tracks multiple distinct entities in a video sequence while managing interactions and occlusions between them.
Single Object Tracking (SOT)
Tracking paradigm where a single predefined object is tracked through a video sequence, typically initialized manually in the first frame.
Siamese Networks
Twin neural network architecture that learns a similarity function to compare a template of the target object with candidate regions in subsequent frames.
Correlation Filters
Tracking approach based on correlation in the frequency domain that maximizes the correlation response between the object model and search regions for efficient localization.
Tracklet
Continuous temporal segment of an object's trajectory, representing a sequence of successive detections belonging to the same entity before merging with other segments.
Re-Identification (Re-ID)
Technique for recognizing the same object after a period of absence or occlusion by comparing its unique appearance characteristics.
Tracking-by-Detection
Tracking strategy that decouples object detection in each frame from the temporal association of detections, usually using pre-trained CNN detectors.
Particle Filter
Sequential Monte Carlo method that represents the probability distribution of the object's state with a set of weighted particles to handle non-linear movements.
Mean Shift
Iterative non-parametric clustering algorithm that finds the mode of a probability distribution to locate the object by following the direction of the highest feature density.
CAMShift
Adaptive extension of Mean Shift that dynamically adjusts the size of the tracking window based on the moments of the object's color distribution.
Online Tracking
Tracking approach that processes frames sequentially and updates trajectories in real-time without access to future frames.
Offline Tracking
Tracking method that optimizes trajectories over the entire complete video, allowing for retroactive correction of association errors.
Feature Matching
Process of identifying correspondences between interest points or visual descriptors from one frame to another to establish tracking continuity.
Hungarian Algorithm
Combinatorial optimization algorithm used to solve the assignment problem between current detections and existing trajectories in multi-object tracking.