Video Object Detection via Frame-Alternating Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in achieving real-time object detection on resource-limited devices such as mobile phones and cameras, due to constraints in memory, computation, and power consumption.
Innovation Solution
The method involves dividing a video into frames and using separate neural networks trained on different domains to generate object information, allowing for efficient object detection by alternating the application of these networks across frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single high-accuracy neural network is used for object detection, then detection accuracy is improved, but computational resource consumption increases
Solution Approach 1:
The patent divides the video processing task into segments by separating frames into different groups (first frames and second frames) and applying different neural networks to each group. This segmentation allows the system to use a computationally intensive high-accuracy network only on necessary frames while using a lighter network on others, thus maintaining detection accuracy while reducing overall computational resource consumption.
2Reliability
If object detection is performed on every video frame, then detection completeness is improved, but processing speed decreases
Solution Approach 1:
The patent implements periodic action by alternately applying different neural networks to different frames in a periodic pattern. The first neural network is applied to first frames while the second neural network is applied to second frames, creating a periodic processing rhythm that balances detection completeness with processing speed requirements.
Solution Approach 2:
The patent applies partial action by using two different neural networks with different levels of computational intensity. The second neural network performs fewer computations and is applied to frames where full detection is less critical, while the first neural network provides more thorough detection on other frames. This partial application of different detection intensities maintains overall detection completeness while improving processing speed.
3Adaptability or versatility
If multiple neural networks are used for different detection tasks, then detection versatility is improved, but system complexity increases
Solution Approach 1:
The patent applies universality by designing a system where two different neural networks work together to handle multiple detection scenarios. Both networks are trained on the same object detection task but with different computational characteristics, allowing the system to adapt to different performance requirements using the same framework. This multi-functional approach enables the system to provide both high-accuracy detection and fast processing without requiring entirely separate systems.
Data Source
AI summary
Techniques and apparatuses enabling high accuracy video object detection using reduced system resource requirements (e.g., reduced computational load, shallower neural network designs, etc.) are described. For example, a search domain of an object detection scheme (e.g., a target object class, a target object size, a target object rotation angle, etc.) may be separated into subdomains (e.g., such as subdomains of object classes, subdomains of object sizes, subdomains object rotation angles, etc.). Specialized, subdomain-level object detection/segmentation tasks may then be separated across sequential video frames. As such, different subdomain-level processing techniques (e.g., via specialized neural networks) may be implemented across different frames of a video sequence. Moreover, redundancy information of consecutive video frames may be leveraged, such that specialized object detection tasks combined with visual object tracking across consecutive frames may enable more efficient (e.g., more accurate, less computationally intensive, etc.) full domain object detection and object segmentation schemes.