Eccentricity Image Fusion for Vehicle Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle operation systems face challenges in accurately and efficiently determining object locations in video data, which is crucial for safe and efficient autonomous or semi-autonomous vehicle navigation, as existing machine vision-based techniques are computationally intensive and less accurate.
Innovation Solution
The method involves combining a red-green-blue (RGB) image with an eccentricity map, based on per-pixel average and variance over a moving window of video frames, and processing this fused data using neural networks like YOLO, Faster-RCNN, or SSD, to improve object location accuracy and reduce computation time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine vision-based techniques are used to determine object locations in video data, then object detection capability is provided, but computation time and resources increase and accuracy decreases
Solution Approach 1:
The patent segments the video processing task into two distinct components: an eccentricity detection stage that identifies regions of interest using motion-based eccentricity maps, and a subsequent object detection stage that processes only these segmented regions. This segmentation reduces the computational burden on the neural network while maintaining detection accuracy, directly resolving the contradiction between computation time and detection capability.
Solution Approach 2:
The patent introduces an intermediary component - the eccentricity map - that acts as a bridge between raw video data and neural network processing. This intermediary pre-processes the video frames to highlight moving objects and their boundaries, enabling the neural network to focus computational resources on relevant regions rather than processing entire frames, thus reducing computation time while improving location accuracy.
2Measurement precision
If machine vision-based techniques are used to determine object locations in video data, then object detection capability is provided, but computation resources increase
Solution Approach 1:
By segmenting the video data into eccentricity maps that highlight only relevant moving regions, the patent reduces the input data volume for the neural network. This segmentation strategy maintains object detection accuracy while significantly reducing the computational resources and energy required for processing, as the neural network operates on condensed, information-dense representations rather than full video frames.
Solution Approach 2:
The patent extracts only the essential information needed for object detection - the eccentricity characteristics of moving objects - from the full video data. This extraction process removes redundant information and computational overhead, allowing the system to achieve accurate object location determination with reduced computation resources and energy consumption.
3Productivity
If existing machine vision techniques are used, then object detection is performed, but accuracy is reduced
Solution Approach 1:
The patent performs preliminary action by generating eccentricity maps before neural network processing. These maps pre-identify moving objects and their boundaries, providing the neural network with prepared, annotated regions that accelerate the detection process. This preliminary preparation maintains high detection speed while simultaneously improving location accuracy by focusing the network's attention on relevant areas.
Solution Approach 2:
The patent changes the parameter representation of video data by transforming standard video frames into eccentricity maps that encode motion-based spatial relationships. This parameter transformation enhances the salience of moving objects and their boundaries, enabling faster and more accurate detection by the neural network, thus resolving the contradiction between detection speed and location precision.
Data Source
AI summary
A system, comprising a computer that includes a processor and a memory, the memory storing instructions executable by the processor to input a red-green-blue (RGB) image and an eccentricity image to a neural network which outputs a located object based on combining the RGB image and the eccentricity image, wherein the eccentricity image is based on a per-pixel rolling average and a per-pixel rolling variance over a moving window of k video frames. The memory can further include instructions executable by the processor to receive the located object at a computing device included in one or more of a vehicle or a traffic information system.