Dynamic Neural Network Selection for IoT Power Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural network (NN) object detection systems in IoT devices face challenges in balancing power consumption and performance, as they require significant computational resources and are often limited by battery power, leading to reduced accuracy and speed under power constraints.
Innovation Solution
A system comprising a video camera, multiple processors, volatile memory, and non-volatile memory that dynamically selects and switches between different neural networks based on performance metrics and operational constraints to optimize power usage and detection accuracy, employing a control layer to manage the object detection process and adjust resources accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single neural network is used for object detection in IoT devices, then the device structure is simple, but the power consumption is high and performance is reduced under battery constraints
Solution Approach 1:
The patent divides the object detection system into multiple neural networks with different complexity levels (first neural network with fewer parameters, second neural network with more parameters). The controller segments the detection task by selecting appropriate networks based on performance requirements and power constraints, allowing the system to balance complexity and energy consumption dynamically.
2Ease of operation
If a single neural network is used for object detection, then the system is simple to operate, but the detection accuracy and speed cannot be optimized under varying power constraints
Solution Approach 1:
The patent implements dynamic selection of neural networks based on real-time performance metrics and operational constraints. The controller automatically switches between the first and second neural networks depending on the required detection accuracy and available power, making the system adaptable without requiring manual intervention or complex user decisions.
3Measurement precision
If a larger neural network with more parameters is used, then the detection accuracy is improved, but the computational resources and power consumption increase
Solution Approach 1:
The patent changes the parameter configuration of neural networks by providing multiple networks with different numbers of parameters (first network with fewer parameters, second network with more parameters). This allows the system to adjust the computational complexity and accuracy trade-off by selecting the appropriate parameter configuration based on current operational requirements and power availability.
4Reliability
If multiple neural networks are deployed, then the performance and power trade-off can be balanced, but the device complexity and memory requirements increase
Solution Approach 1:
The patent makes the object detection system multi-functional by enabling it to operate in different performance modes using multiple neural networks. The same hardware platform can adapt to varying performance requirements and power constraints without needing separate dedicated systems, achieving universality through configurable neural network selection rather than physical complexity.
Data Source
AI summary
This disclosure relates to an apparatus for object detection. The apparatus comprises a video camera, an object detector, and a controller. The video camera may be configured to generate a video stream of frames. The object detector may be configured to accept the video stream as input data and to perform object detection. The controller may be coupled to the video camera and the object detector. The controller may be configured to manage object detection in order to satisfy a performance metric and/or operate within an operational constraint.