Recurrent Neural Feature Extraction for Accurate Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods require significant processing resources and increased model complexity to achieve high accuracy, especially for hard-to-recognize objects, leading to inefficiencies in computational power, memory, and energy usage.
Innovation Solution
A method utilizing both local and common feature extraction through a neural network with recurrent behavior, combining spatial and temporal aspects, to enhance object detection accuracy without excessive resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing object detection methods are used to achieve high accuracy for hard-to-recognize objects, then detection accuracy is improved, but processing resources and model complexity increase significantly
Solution Approach 1:
The patent divides the image processing into multiple data streams, where each stream focuses on detecting a specific aspect or type of object. This segmentation allows the system to achieve high accuracy for specific object types without requiring a single complex model to handle all detection tasks, thereby reducing overall model complexity while maintaining or improving detection accuracy for target objects.
Solution Approach 2:
The patent introduces a temporal dimension by processing image data across multiple frames rather than analyzing single images in isolation. By leveraging temporal information from sequential frames, the system can detect hard-to-recognize objects more accurately without significantly increasing spatial model complexity, as the additional dimension provides complementary information that aids detection.
2Measurement precision
If existing object detection methods are used to achieve high accuracy for hard-to-recognize objects, then detection accuracy is improved, but computational power and memory usage increase
Solution Approach 1:
By segmenting the detection task into multiple specialized data streams, each stream processes only relevant features for its specific detection goal. This reduces the computational burden on each individual stream compared to a monolithic model processing all features, leading to lower overall computational power consumption while maintaining high detection accuracy through the coordinated output of multiple streams.
Solution Approach 2:
The patent processes only the necessary portions of image data in each data stream, focusing computational resources on extracting relevant features for specific object types rather than performing exhaustive analysis on all image content. This partial action approach reduces unnecessary computational operations while maintaining detection accuracy for target objects.
3Measurement precision
If existing object detection methods are used to achieve high accuracy for hard-to-recognize objects, then detection accuracy is improved, but energy consumption increases
Solution Approach 1:
The energy-consuming detection task is divided into multiple data streams, each handling a specific aspect of object detection. This segmentation distributes the energy load across multiple lighter processing tasks rather than one heavy task, reducing peak energy consumption while achieving high accuracy through the combined results of all streams.
Solution Approach 2:
The patent leverages periodic action by processing image data in sequential frames over time, allowing the system to accumulate detection information across multiple time periods. This temporal approach enables more accurate detection of hard-to-recognize objects without requiring proportionally higher energy consumption in each individual frame, as the detection benefit accumulates over the periodic processing of multiple frames.
Data Source
Figure 1A~2B
Figure 3
Figure 4
AI summary
An object detection arrangement (100) comprising a controller (101) configured to: a) receive (212) a plurality of image data streams; b) perform feature extraction on each of the received a plurality of images providing a plurality of feature data streams; and to c) perform a common feature extraction based on the plurality of feature data streams providing as common feature data stream for object detection.