Glossary

9.3 Object tracking algorithms

12 min read•august 21, 2024

Object tracking algorithms are essential in computer vision, enabling continuous localization of objects across video frames. These algorithms play a crucial role in applications like surveillance, , and human-computer interaction, requiring robust solutions to handle challenges such as occlusions and appearance changes.

This topic covers the fundamentals of object tracking, including its definition, applications, and challenges. It explores various types of tracking methods, feature selection techniques, and algorithms for both single and . The notes also delve into approaches, performance evaluation, and real-time considerations in object tracking.

Fundamentals of object tracking

  • Object tracking forms a crucial component of computer vision systems enabling the continuous localization of objects across video frames
  • Plays a vital role in various applications including surveillance, autonomous vehicles, and human-computer interaction
  • Requires robust algorithms to handle challenges like occlusions, appearance changes, and complex motion patterns

Definition and purpose

Top images from around the web for Definition and purpose

  • Frontiers | Event-Based Trajectory Prediction Using Spiking Neural Networks View original

    Is this image relevant?

  • Frontiers | UAV-UGV-UMV Multi-Swarms for Cooperative Surveillance View original

    Is this image relevant?

  • Frontiers | Event-Based Trajectory Prediction Using Spiking Neural Networks View original

    Is this image relevant?

1 of 3

Top images from around the web for Definition and purpose

  • Frontiers | Event-Based Trajectory Prediction Using Spiking Neural Networks View original

    Is this image relevant?

  • Frontiers | UAV-UGV-UMV Multi-Swarms for Cooperative Surveillance View original

    Is this image relevant?

  • Frontiers | Event-Based Trajectory Prediction Using Spiking Neural Networks View original

    Is this image relevant?

1 of 3
  • Automated process of locating a moving object (or multiple objects) over time using a camera
  • Aims to generate the trajectory of an object by locating its position in every frame of the video
  • Involves two key steps: object detection and object association between frames
  • Enables analysis of object behavior, prediction of future locations, and understanding of scene dynamics

Applications in computer vision

  • systems use tracking to monitor suspicious activities and ensure security
  • Autonomous vehicles rely on object tracking to navigate safely and avoid collisions
  • Human-computer interaction utilizes tracking for gesture recognition and augmented reality experiences
  • Sports analytics employs tracking to analyze player movements and team strategies
  • Medical imaging benefits from tracking for monitoring organ motion during procedures

Challenges in object tracking

  • Occlusions occur when objects become partially or fully hidden behind other objects
  • Illumination changes affect the appearance of objects, making consistent tracking difficult
  • Complex object motions, including abrupt changes in direction or speed, challenge tracking algorithms
  • Scale variations as objects move closer or farther from the camera impact
  • Background clutter can confuse trackers, especially when objects have similar appearances to the background

Types of object tracking

  • Object tracking methods can be broadly categorized into three main types based on their approach
  • Each type has its strengths and weaknesses, making them suitable for different scenarios and applications
  • Understanding these types helps in selecting the most appropriate tracking method for specific use cases

Point tracking

  • Represents objects as points and tracks their positions across frames
  • Utilizes past object positions to predict future locations
  • Suitable for tracking small objects or when detailed shape information is not required
  • and are common algorithms used in point tracking
  • Challenges include handling occlusions and distinguishing between multiple similar objects

Kernel tracking

  • Represents objects using shapes like rectangles or ellipses (kernels)
  • Tracks objects by computing the motion of the kernel in consecutive frames
  • Mean-shift and KLT (Kanade-Lucas-Tomasi) trackers are popular kernel-based methods
  • Effective for tracking objects with consistent appearance and gradual motion
  • Struggles with significant scale changes or rotations of the object

Silhouette tracking

  • Tracks objects by estimating their complete region in each frame
  • Utilizes the encoded information inside the object region for tracking
  • Contour tracking and shape matching are common approaches in silhouette tracking
  • Provides detailed information about object shape and deformation
  • Well-suited for tracking non-rigid objects with complex shapes
  • Computationally intensive compared to point and kernel tracking methods

Feature selection for tracking

  • Feature selection plays a crucial role in the performance and of object tracking algorithms
  • Choosing appropriate features enables trackers to distinguish objects from backgrounds and handle various challenges
  • Different types of features capture distinct aspects of objects, often combined for more robust tracking

Color features

  • Represent objects using color information from various color spaces (RGB, HSV, LAB)
  • Color histograms capture the distribution of colors within an object region
  • Robust to rotation and partial occlusions but sensitive to illumination changes
  • Color-based tracking performs well for objects with distinct colors from the background
  • Challenges include tracking objects with similar colors to the background or under varying lighting conditions

Shape features

  • Describe the geometric properties and contours of objects
  • Include features like edges, corners, and shape descriptors (Hu moments, Fourier descriptors)
  • Invariant to illumination changes and can handle partial occlusions
  • Effective for tracking rigid objects with well-defined shapes
  • May struggle with deformable objects or those with complex, changing shapes

Motion features

  • Capture the dynamic characteristics of moving objects
  • Optical flow estimates pixel-wise motion between consecutive frames
  • Motion history images represent the recent motion of objects
  • Useful for distinguishing between moving objects and static backgrounds
  • Challenges arise with camera motion or when tracking slow-moving objects

Texture features

  • Describe the spatial arrangement of intensities or colors in object regions
  • Include features like Local Binary Patterns (LBP) and Gabor filters
  • Robust to illumination changes and can differentiate objects with similar colors
  • Effective for tracking objects with distinct textural patterns
  • May struggle with smooth or uniform objects lacking texture

Single object tracking algorithms

  • focuses on following a single target throughout a video sequence
  • These algorithms form the foundation for more complex multi-object tracking systems
  • Each method has its strengths and is suited for different tracking scenarios

Mean-shift tracking

  • Iterative algorithm that finds the mode of a probability distribution
  • Locates the peak of a confidence map derived from object features (often color histograms)
  • Efficient and works well for objects with distinct color distributions
  • Adapts to partial occlusions and gradual appearance changes
  • Limitations include difficulty handling full occlusions and significant scale changes

Kalman filter

  • Recursive estimator that predicts object state based on previous measurements and a motion model
  • Consists of two steps: prediction and update
  • Optimal for linear systems with Gaussian noise
  • Widely used in point tracking and works well for objects with predictable motion
  • Struggles with non-linear motion and multi-modal distributions

Particle filter

  • Monte Carlo method that represents the posterior distribution of object states with a set of weighted samples (particles)
  • Capable of handling non-linear motion and multi-modal distributions
  • Adapts well to complex scenarios and can recover from temporary tracking failures
  • Computationally more intensive than Kalman filter, especially with a large number of particles
  • Performance depends on the number of particles and the quality of the motion model

Optical flow

  • Estimates the apparent motion of objects between consecutive frames
  • Dense optical flow computes motion for every pixel, while sparse optical flow tracks specific features
  • Lucas-Kanade and Horn-Schunck are popular optical flow algorithms
  • Effective for tracking objects with texture and gradual motion
  • Challenges include handling large displacements and areas with uniform intensity

Multiple object tracking

  • Extends single object tracking to simultaneously track multiple objects in a scene
  • Crucial for applications like crowd analysis, sports analytics, and traffic monitoring
  • Introduces additional challenges such as object interactions and identity management

Data association methods

  • Techniques to match detected objects with existing tracks across frames
  • Include methods like Hungarian algorithm and Global Nearest Neighbor (GNN)
  • Solve the assignment problem to minimize the overall cost of object-to-track associations
  • Challenges arise with occlusions, object entries/exits, and similar-looking objects
  • Performance depends on the quality of object detection and

Joint probabilistic data association

  • Probabilistic approach that considers all possible associations between measurements and tracks
  • Computes association probabilities for each measurement-track pair
  • Updates track states using weighted combinations of all measurements
  • Handles uncertain associations and performs well in cluttered environments
  • Computationally intensive for scenarios with many objects and measurements

Multiple hypothesis tracking

  • Maintains multiple hypotheses about object-to-track associations over time
  • Defers making hard decisions when associations are ambiguous
  • Prunes unlikely hypotheses to manage computational complexity
  • Effective for handling complex scenarios with frequent occlusions and crossovers
  • Requires careful parameter tuning to balance between maintaining hypotheses and computational efficiency

Deep learning in object tracking

  • Leverages the power of deep neural networks to learn robust feature representations for tracking
  • Has significantly improved tracking performance in challenging scenarios
  • Enables end-to-end learning of tracking algorithms, reducing the need for hand-crafted features

Convolutional neural networks

  • Extract hierarchical features from input images, capturing both low-level and high-level object characteristics
  • Pre-trained CNNs (VGG, ResNet) often used as feature extractors in tracking frameworks
  • Siamese CNN architectures compare object templates with search regions for localization
  • Challenges include adapting to appearance changes and maintaining real-time performance
  • Transfer learning techniques help in applying CNNs to specific tracking domains

Siamese networks

  • Consist of two identical subnetworks that process the object template and search region
  • Learn a similarity function to compare the template with candidate regions in new frames
  • Fully-convolutional Siamese networks enable efficient sliding-window search
  • Perform well in short-term tracking scenarios and can handle previously unseen object classes
  • Limitations include difficulty in adapting to significant appearance changes over long sequences

Recurrent neural networks

  • Model temporal dependencies in object motion and appearance changes
  • Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) are common RNN variants used in tracking
  • Can learn to predict object locations and update appearance models over time
  • Effective for long-term tracking and handling temporary occlusions
  • Challenges include training on long sequences and balancing memory requirements

Performance evaluation

  • Critical for comparing different tracking algorithms and assessing their strengths and weaknesses
  • Helps in selecting appropriate trackers for specific applications and identifying areas for improvement
  • Standardized evaluation protocols enable fair comparisons across different research works

Metrics for tracking accuracy

  • measures the overlap between predicted and ground truth bounding boxes
  • Center error calculates the distance between predicted and ground truth object centers
  • Success rate plots show the percentage of frames with IoU above varying thresholds
  • Precision plots display the percentage of frames with center error below different thresholds
  • Multiple Object Tracking (MOTA) combines errors from false positives, misses, and identity switches

Benchmark datasets

  • provides a diverse set of sequences for single object tracking
  • benchmark focuses on pedestrian tracking in crowded scenes
  • VOT (Visual Object Tracking) challenge offers sequences with various tracking difficulties
  • dataset for long-term tracking evaluation
  • specifically for drone-based tracking scenarios

Challenges and competitions

  • VOT Challenge annually evaluates state-of-the-art trackers on a new dataset
  • MOT Challenge provides a platform for evaluating multiple object tracking algorithms
  • CVPR tracking challenges focus on specific aspects like long-term tracking or 3D tracking
  • Real-time requirements in some competitions push for efficient algorithm implementations
  • Encourage development of robust trackers capable of handling diverse real-world scenarios

Real-time considerations

  • Many tracking applications require real-time performance for practical deployment
  • Balancing accuracy and speed is crucial for developing effective tracking systems
  • Real-time tracking enables immediate decision-making in applications like autonomous driving and surveillance

Computational efficiency

  • Algorithmic optimizations reduce computational complexity without sacrificing accuracy
  • Efficient feature extraction methods (integral images, approximated filters) speed up processing
  • Selective search strategies limit the region of interest for object localization
  • Parallel processing techniques exploit multi-core CPUs for faster computations
  • Incremental learning approaches update models efficiently without full retraining

Hardware acceleration

  • Graphics Processing Units (GPUs) significantly accelerate deep learning-based trackers
  • Field-Programmable Gate Arrays (FPGAs) offer low- solutions for embedded systems
  • Tensor Processing Units (TPUs) provide specialized hardware for neural network computations
  • Mobile GPUs enable real-time tracking on smartphones and tablets
  • Distributed computing systems allow processing of multiple video streams simultaneously

Trade-offs in accuracy vs speed

  • Reducing the number of particles in particle filters decreases accuracy but improves speed
  • Lowering the resolution of input images speeds up processing at the cost of fine-grained tracking
  • Simplified network architectures in deep learning trackers offer faster inference with potential accuracy loss
  • Frame skipping techniques reduce computational load but may miss rapid object movements
  • Online model update frequency affects the tracker's adaptability and computational requirements

Handling occlusions

  • Occlusions present significant challenges in object tracking, often leading to tracking failures
  • Robust handling of occlusions is crucial for maintaining long-term tracking performance
  • Different strategies are employed based on the nature and duration of occlusions

Partial occlusion strategies

  • Part-based models represent objects as collections of parts, allowing tracking of visible portions
  • Adaptive appearance models update object representations to focus on unoccluded regions
  • detection mechanisms identify partially occluded areas and adjust tracking accordingly
  • Spatial-temporal context information helps in predicting object locations during partial occlusions
  • Multiple feature fusion improves robustness by relying on features less affected by occlusions

Full occlusion recovery

  • Short-term prediction methods estimate object locations during brief full occlusions
  • Re-detection strategies search for the object in a wider area after occlusion ends
  • Appearance model preservation maintains object representations during occlusion periods
  • Multi-camera setups provide alternative views to track objects through occlusions
  • Long-term memory mechanisms store object information for extended periods to aid in re-identification

Long-term tracking

  • Combines short-term tracking with detection-based re-identification for handling long occlusions
  • Implements confidence measures to determine when to switch between tracking and detection modes
  • Utilizes scene context and object interaction models to predict reappearance locations
  • Employs efficient object instance search techniques for fast re-detection in large search areas
  • Incorporates online learning to adapt to appearance changes over extended tracking periods

Object tracking in complex scenes

  • Real-world tracking scenarios often involve challenging environmental conditions
  • Robust trackers must adapt to these complexities to maintain accurate and consistent tracking
  • Understanding and addressing these challenges is crucial for developing practical tracking systems

Varying illumination conditions

  • Illumination-invariant features (e.g., gradient-based descriptors) reduce sensitivity to lighting changes
  • Adaptive histogram equalization techniques normalize object appearance across different lighting conditions
  • Multi-spectral imaging (infrared, thermal) provides additional information in low-light scenarios
  • Online appearance model updates allow trackers to adapt to gradual illumination changes
  • Shadow detection and removal methods improve tracking accuracy in outdoor environments

Dynamic backgrounds

  • Background subtraction techniques identify moving objects in changing scenes
  • Optical flow analysis distinguishes between object motion and background motion
  • Adaptive background models account for repetitive background movements (waving trees, flowing water)
  • Region-based tracking methods focus on object appearance, reducing reliance on static backgrounds
  • Semantic segmentation helps in differentiating between objects and dynamic background elements

Camera motion compensation

  • Estimate global motion using techniques like homography estimation or RANSAC
  • Apply motion compensation to stabilize the video sequence before tracking
  • Ego-motion estimation in moving cameras (vehicles, drones) to separate object motion from camera motion
  • Visual odometry techniques track camera movement in 3D space for accurate motion compensation
  • Inertial measurement unit (IMU) data fusion improves motion estimation in rapidly moving cameras
  • Object tracking continues to evolve with advancements in computer vision and machine learning
  • Emerging technologies and novel approaches promise to address current limitations and enable new applications
  • Research in these areas aims to make tracking systems more robust, efficient, and adaptable to diverse scenarios

Multi-modal fusion

  • Integrates data from multiple sensors (RGB cameras, depth sensors, LiDAR) for more comprehensive tracking
  • Combines visual information with other modalities like audio or thermal imaging
  • Exploits complementary strengths of different modalities to improve tracking in challenging conditions
  • Develops fusion strategies that handle asynchronous and heterogeneous data streams
  • Enables robust tracking in scenarios where single modalities fail (e.g., visual tracking in darkness)

Online learning and adaptation

  • Continuous learning approaches allow trackers to adapt to changing object appearances and environments
  • Meta-learning techniques enable quick adaptation to new objects with minimal training data
  • Self-supervised learning methods leverage unlabeled video data to improve tracking models
  • Incremental learning strategies update deep neural networks efficiently during tracking
  • Addresses the challenge of long-term tracking in dynamic and evolving scenes

Edge computing for tracking

  • Distributes tracking computations between edge devices and cloud infrastructure
  • Enables real-time tracking on resource-constrained devices (smartphones, IoT sensors)
  • Develops lightweight tracking algorithms optimized for edge deployment
  • Explores federated learning approaches for collaborative model improvement across multiple edge devices
  • Addresses privacy concerns by processing sensitive tracking data locally on edge devices

Key Terms to Review (26)

Accuracy: Accuracy refers to the degree to which a measurement, classification, or prediction corresponds to the true value or outcome. In various applications, especially in machine learning and computer vision, accuracy is a critical metric for assessing the performance of models and algorithms, indicating how often they correctly identify or classify data.
Autonomous vehicles: Autonomous vehicles are self-driving cars or systems that can navigate and operate without human intervention, utilizing a combination of sensors, cameras, and advanced algorithms. These vehicles rely on real-time data processing to understand their environment, make decisions, and safely transport passengers or goods. This technology is crucial for applications like smart transportation systems, reducing traffic accidents, and enhancing mobility.
Deep Learning: Deep learning is a subset of machine learning that uses neural networks with many layers to analyze data patterns and make predictions. It excels in handling complex data types such as images and video, enabling advanced capabilities in areas like object tracking and autonomous systems. By mimicking the way the human brain processes information, deep learning allows for significant advancements in recognition, classification, and decision-making tasks.
Feature extraction: Feature extraction is the process of transforming raw data into a set of characteristics or features that can effectively represent the underlying structure of the data for tasks such as classification, segmentation, or recognition. This process is crucial in various applications where understanding and identifying relevant patterns from complex data is essential, enabling more efficient algorithms to work with less noise and improved performance.
Intersection over Union (IoU): Intersection over Union (IoU) is a metric used to evaluate the accuracy of an object detection model by measuring the overlap between the predicted bounding box and the ground truth bounding box. This ratio is calculated by dividing the area of overlap between the two boxes by the area of their union, providing a single value that ranges from 0 to 1, where a value of 1 indicates perfect overlap. This metric is crucial for assessing performance in tasks such as object detection, tracking, and segmentation.
Kalman Filter: The Kalman filter is an algorithm that provides estimates of unknown variables over time using a series of measurements observed over time, which contain noise and other inaccuracies. It is widely used for object tracking, filtering out noise from sensor data, and making predictions about future states based on current observations. This makes it particularly useful in applications involving dynamic systems where tracking and estimating the state of moving objects is essential.
KITTI Dataset: The KITTI Dataset is a large-scale collection of image data and associated annotations specifically designed for evaluating computer vision algorithms, particularly in the realm of autonomous driving. It contains real-world images captured from a moving vehicle in urban and rural environments, providing a rich source of data for developing and testing object tracking algorithms and other vision-related tasks.
Lasot (large-scale single object tracking): Large-scale single object tracking (LASOT) refers to the process of continuously monitoring and following a specific object across various frames in a video or image sequence, especially when the object is large and may experience significant changes in scale, appearance, or motion. This technique is crucial for applications that require robust tracking performance over long durations, such as in surveillance, robotics, and autonomous vehicles. Effective LASOT algorithms balance computational efficiency with accuracy to handle real-time processing demands.
Latency: Latency refers to the delay or lag in processing time that occurs in systems, particularly in the context of tracking objects in real-time scenarios. It is a critical factor that impacts the responsiveness and efficiency of object tracking algorithms, as lower latency allows for quicker updates and better tracking accuracy. In high-speed environments, minimizing latency is essential to maintain the integrity of data and ensure accurate object movement representation.
Mean Shift: Mean shift is a non-parametric clustering algorithm used in computer vision and image processing for segmenting data points into clusters based on their density. It works by iteratively shifting points towards the mean of the data points within a specified radius, effectively finding dense regions in the feature space. This approach allows for flexible and adaptive clustering, which is particularly useful for separating objects from backgrounds or tracking moving objects over time.
Mot (multiple object tracking): Multiple object tracking (MOT) is a computer vision task that focuses on identifying and following multiple objects over time in a sequence of frames. This involves not only detecting these objects but also maintaining their identities as they move through different frames, which is crucial for applications such as surveillance, robotics, and autonomous driving. Effective MOT algorithms utilize various techniques to manage occlusions, object interactions, and changes in appearance, ensuring accurate tracking in dynamic environments.
Motchallenge: Motchallenge is a benchmark dataset and evaluation framework for assessing the performance of object tracking algorithms in computer vision. It provides a standardized set of sequences with ground truth data that allows researchers to compare the effectiveness of different tracking methods under various conditions such as illumination changes, occlusions, and scale variations.
Multiple object tracking: Multiple object tracking is the process of detecting and following multiple objects as they move through a scene over time. This involves maintaining identities for each object, predicting their positions, and managing occlusions or interactions between them. Effective multiple object tracking combines both detection and tracking algorithms to provide accurate position updates for each object within a video sequence.
Occlusion: Occlusion refers to the phenomenon where an object in a visual scene is partially or completely hidden by another object. This effect can complicate the understanding of motion and depth in visual perception, making it essential for algorithms to account for occlusions when analyzing moving objects or tracking them over time.
Otb (object tracking benchmark): The Object Tracking Benchmark (OTB) is a comprehensive evaluation framework designed to assess the performance of object tracking algorithms in various scenarios. It provides a standardized set of sequences, metrics, and protocols to facilitate fair comparisons among different tracking methods, enabling researchers to benchmark their algorithms and identify strengths and weaknesses. By systematically evaluating object tracking algorithms, OTB aids in advancing the field and improving tracking accuracy across diverse applications.
Particle filter: A particle filter is a computational method used for estimating the state of a dynamic system through a set of weighted samples, known as particles. This technique is particularly effective in object tracking, as it can handle non-linear and non-Gaussian models by representing the posterior distribution of the system's state with a collection of particles that are updated over time based on observed measurements.
Precision-Recall: Precision-recall is a performance metric used to evaluate the effectiveness of classification models, particularly in situations with imbalanced classes. Precision measures the accuracy of positive predictions, while recall (or sensitivity) assesses how well a model identifies actual positives. These metrics are crucial for understanding the trade-offs between false positives and false negatives in various applications, especially in visual recognition and tracking tasks.
Real-time processing: Real-time processing refers to the ability of a system to process data and provide immediate output or response without any noticeable delay. This capability is crucial in various applications, as it ensures that data is analyzed and acted upon instantly, which is especially important in situations requiring quick decision-making. The effectiveness of real-time processing can be seen in various fields, including image manipulation, feature detection, tracking moving objects, and enabling autonomous systems to navigate and react to their environments seamlessly.
Robustness: Robustness refers to the ability of a system or algorithm to maintain performance and provide accurate results despite variations in input data or environmental conditions. In the context of feature detection and tracking, robustness is crucial as it determines how well these algorithms perform under different scenarios such as changes in scale, rotation, lighting, and occlusion.
Scalability: Scalability refers to the capability of a system to handle a growing amount of work or its potential to accommodate growth. In the context of object tracking algorithms, scalability is crucial as it determines how well these algorithms can perform when faced with increasing numbers of objects, higher resolutions, or more complex scenarios without a significant drop in performance or accuracy.
Single Object Tracking: Single object tracking is a computer vision task focused on following a specific object as it moves through a sequence of frames in a video or a series of images. This technique is crucial for applications such as surveillance, human-computer interaction, and autonomous vehicles, as it allows systems to maintain continuous observation of the target object, even as it undergoes changes in appearance, scale, or orientation.
Track loss: Track loss refers to the failure of an object tracking algorithm to maintain the identity or continuous tracking of a target object over time. This can happen due to various factors, such as occlusions, changes in appearance, or abrupt movements. Understanding track loss is crucial for improving the robustness and reliability of object tracking algorithms.
Tracking accuracy: Tracking accuracy refers to the degree to which an object tracking algorithm correctly identifies and follows the position of a moving object across successive frames in a video sequence. It is crucial for evaluating the performance of tracking algorithms, as high accuracy ensures that the tracked object is consistently represented, minimizing errors in position and trajectory estimation. This term is essential for applications such as surveillance, robotics, and autonomous vehicles, where precise tracking directly impacts functionality and safety.
Uav123 dataset: The uav123 dataset is a large-scale benchmark designed specifically for evaluating visual object tracking algorithms in aerial video sequences. It contains 123 video sequences captured from UAVs (unmanned aerial vehicles) in diverse environments, which makes it a critical resource for researchers working on object tracking algorithms, especially in challenging scenarios like occlusions, illumination changes, and varying object scales.
Video surveillance: Video surveillance is the use of video cameras to monitor activities in specific areas for security and safety purposes. It involves capturing, recording, and analyzing video footage to detect and prevent crimes or unauthorized activities, making it a critical component of modern security systems.
Visual Object Tracking (VOT): Visual Object Tracking is the process of locating and following an object of interest in a sequence of video frames. This technique is essential for various applications, such as surveillance, autonomous vehicles, and human-computer interaction. VOT involves several challenges, including variations in scale, occlusion, and changes in the object's appearance over time.
© 2024 Fiveable Inc. All rights reserved.
AP® and SAT® are trademarks registered by the College Board, which is not affiliated with, and does not endorse this website.
Back
Practice QuizGlossary
Practice QuizGlossary
Next guide