Movement Path Inference Using Shape and Position Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis techniques struggle to correctly recognize movement types when multiple learned movement path curves have similar shapes, leading to incorrect recognition due to positional differences.
Innovation Solution
An inference apparatus that computes shape and position similarities between learned and observed movement path curves, utilizing speed and distribution attributes of curve components to enhance recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network with a large number of parameters is used to improve AI performance, then processing capability and accuracy are improved, but the number of floating point operations increases leading to higher energy consumption and longer processing time
Solution Approach 1:
The patent segments the neural network into multiple hierarchical levels (first level, second level, third level) with different quantization precision settings. High-precision processing is applied only where critical for accuracy, while lower precision is used in other regions. This segmentation allows the system to maintain AI processing accuracy while reducing overall computational complexity and inference time.
Solution Approach 2:
The patent applies different quantization precision levels to different parts of the neural network based on their specific requirements. Critical layers maintain higher precision (e.g., 8-bit or higher) to preserve accuracy, while non-critical layers use lower precision (e.g., 4-bit or lower) to reduce computational load. This local quality differentiation resolves the contradiction by optimizing the balance between accuracy and speed in different network regions.
2Measurement precision
If a neural network with a large number of parameters is used to improve AI performance, then processing capability is improved, but the number of floating point operations increases leading to higher energy consumption
Solution Approach 1:
The patent segments the neural network into multiple hierarchical levels with different quantization precision settings. By dividing the network into regions requiring high precision and regions where lower precision suffices, the system maintains accuracy where needed while dramatically reducing the total number of floating point operations and associated energy consumption in less critical areas.
Solution Approach 2:
The patent implements local quality differentiation through heterogeneous quantization, where different portions of the neural network operate at different precision levels. This allows energy-efficient low-precision processing in non-critical regions while maintaining high precision only where necessary for accurate AI inference, thereby resolving the energy consumption vs. accuracy contradiction.
3Speed
If quantization precision is reduced to decrease memory bandwidth requirements and improve processing speed, then inference speed is improved, but processing accuracy deteriorates
Solution Approach 1:
The patent segments the neural network into multiple levels with different quantization precision assignments. This segmentation enables the system to achieve high inference speeds in layers where low precision is sufficient, while maintaining adequate accuracy in layers where higher precision is required, thus resolving the speed-accuracy trade-off through spatial differentiation of precision requirements.
Solution Approach 2:
The patent applies local quality optimization by assigning different quantization precision levels to different neural network layers based on their specific functional requirements. Critical layers maintain higher precision to preserve accuracy, while non-critical layers use lower precision to maximize processing speed, achieving an optimal balance between speed and accuracy across the entire network.
4Productivity
If heterogeneous computing units with different precision processing capabilities are used to improve processing efficiency, then energy efficiency is improved, but device complexity increases
Solution Approach 1:
The patent segments the computing system into multiple processing units with different precision capabilities, organized in a hierarchical structure. This segmentation allows the system to distribute computational tasks across units of appropriate precision levels, improving overall processing efficiency while managing complexity through structured organization and unified management of heterogeneous resources.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A shape similarity computation unit (2021) computes, as a shape similarity, a similarity in shape between a learning movement path curve which is a movement path curve obtained through learning and an observation movement path curve which is a movement path curve obtained through observation. A position similarity computation unit (2022) computes, as a position similarity, a similarity in position between the learning movement path curve and the observation movement path curve when the learning movement path curve and the observation movement path curve are placed in a same coordinate space. A conformity computation unit (203) computes a conformity between the learning movement path curve and the observation movement path curve, using the shape similarity and the position similarity.