Automated Deep Object Tracking Dataset Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep neural network (DNN) tracking algorithms face challenges in accurately tracking multiple objects due to the lack of annotated datasets with temporal characteristics, as manual annotation is costly and difficult, especially when applying different track management schemes.
Innovation Solution
The method automates the annotation of datasets by correlating object features across frames using likelihood functions to assign target IDs and track types, enabling the application of various track management schemes without human intervention, resulting in an annotated dataset suitable for training DNNs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation is used to assign target IDs and track types, then the dataset contains accurate temporal characteristics, but the annotation cost and time increase significantly
Solution Approach 1:
The system performs self-annotation by automatically assigning target IDs and track types to objects in video frames using algorithms that analyze object features, positions, and motion patterns. The annotation system serves itself without human intervention, generating temporally characteristic datasets automatically through computational processes that evaluate likelihood functions and apply track management schemes.
Solution Approach 2:
The patent replaces the mechanical manual annotation process with an automated computational system. Instead of human annotators manually assigning IDs and track types, the system uses algorithms that calculate likelihood functions, compare object features across frames, and automatically apply track management rules to generate annotations.
2Measurement precision
If manual annotation with high detail is performed, then temporal characteristics are accurately captured, but the complexity of following track management rules increases difficulty
Solution Approach 1:
The system automatically applies track management schemes without human intervention. The annotation system evaluates object features, calculates likelihood functions, and determines target IDs and track types through computational processes that inherently follow track management rules, eliminating the need for human annotators to understand or follow complex rules.
Solution Approach 2:
The patent replaces the complex manual process of following track management rules with an automated algorithmic system. The computational approach systematically evaluates objects against track management criteria through likelihood calculations and feature comparisons, making the process as easy as executing code rather than manually interpreting and applying complex rules.
3Reliability
If datasets are annotated with temporal characteristics, then deep tracking performance improves, but the quantity of manually annotated datasets remains limited
Solution Approach 1:
The system generates annotated datasets with temporal characteristics automatically through self-service computation. By implementing automated algorithms that assign target IDs and track types across video frames, the system can process large volumes of data without human intervention, dramatically increasing dataset production capacity while maintaining the temporal annotations necessary for reliable deep tracking performance.
Solution Approach 2:
The patent replaces manual dataset annotation with automated computational processes. The system uses algorithms to efficiently process video data, calculate likelihood functions, and generate temporally characteristic annotations at scale, transforming dataset production from a labor-intensive bottleneck to an automated high-throughput process that can supply sufficient training data for deep tracking models.
Data Source
AI summary
Methods and systems for generating an annotated dataset for training a deep tracking neural network, and training of the neural network using the annotated dataset. For each object in each frame of a dataset, one or more likelihood functions are calculated to correlate feature score of the object with respective feature scores each associated with one or more previously assigned target identifiers (IDs) in a selected range of frames. A target ID is assigned to the object by assigning a previously assigned target ID associated with a calculated highest likelihood or assigning a new target ID. Track management is performed according to a predefined track management scheme to assign a track type to the object. This is performed for all objects in all frames of the dataset. The resulting annotated dataset contains target IDs and track types assigned to all objects in all frames.