Activity Detection Device Using Trainable Component Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing activity detection systems face limitations in accuracy due to fixed system architectures and the need for large amounts of learning data, leading to incomplete training and labor-intensive handling of new objects, with feature extraction focusing primarily on convolutional neural networks without considering spatial positional relationships.
Innovation Solution
An activity detection system utilizing individually trainable component models, such as optical flow, object tracking, and object skeleton models, to generate characteristic information and object states within a Markov model, enabling flexible and accurate activity detection without requiring extensive retraining of the entire system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a fixed system architecture using convolutional neural network is used for activity detection, then the system structure is simple, but the detection accuracy is limited and cannot flexibly adapt to different objects
Solution Approach 1:
The system is divided into multiple independently trainable component models (object detection model, object tracking model, object skeleton model, optical flow model, object attribute model) that can be trained separately and then integrated. This segmentation allows each component to be optimized independently while maintaining overall system flexibility.
Solution Approach 2:
The system architecture transitions from a fixed structure to a dynamic, adaptable structure where individual component models can be selectively trained, updated, and retrained based on specific needs. This enables the system to adapt to different objects and scenarios without requiring complete system retraining.
2Measurement precision
If the entire system is retrained to handle new objects, then the detection accuracy for new objects improves, but the time and computational resources required increase significantly
Solution Approach 1:
By segmenting the system into independently trainable component models, only the specific component related to the new object needs to be retrained rather than the entire system. This dramatically reduces retraining time and computational resources while maintaining detection accuracy.
Solution Approach 2:
The component models are pre-trained on general object data, establishing a foundation that can be quickly adapted to new objects through selective retraining. This preliminary training reduces the time and data required for adapting to new objects.
3Loss of information
If only convolutional neural network is used for feature extraction, then the system is simple to implement, but semantic information such as spatial positional relationships and orientations are not considered
Solution Approach 1:
Multiple component models (object detection, tracking, skeleton, optical flow, attribute models) are combined to extract comprehensive features including spatial positional relationships, orientations, and other semantic information that a single CNN cannot capture alone.
Solution Approach 2:
Each component model serves multiple functions: object detection model identifies objects, tracking model provides temporal continuity, skeleton model extracts spatial relationships, optical flow model captures motion, and attribute model provides semantic characteristics. This multi-functional approach comprehensively retains semantic information.
Data Source
AI summary
An object of the disclosure is to provide flexible and highly accurate activity detection means. Provided is an activity detection device including: an input unit that inputs an image sequence including a first image and a second image; an object detection unit that detects a first object in the image sequence; a component model unit that generates first characteristic information characterizing the first object and includes at least one individually trainable component model; and an activity detection unit that generates a first object state corresponding to the first object in the first image and a second object state corresponding to the first object in the second image based on the first characteristic information and determines an activity related to the first object based on the first and second object states.