Object Detection via Image Feature Maps and Saliency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection devices struggle to accurately detect objects located far from the camera due to the small absolute value of movement vectors in optical flows, leading to inaccurate differences with optical flows from object-free images.
Innovation Solution
An object detection device that generates image pyramids by reducing frame images, extracts salient pixels based on brightness thresholds, and integrates them into image feature maps to represent object likelihood distributions, enabling accurate detection from near to distant ranges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If optical flow based on images of distant objects is used, then the detection range is extended, but the detection accuracy deteriorates due to small movement vector values
Solution Approach 1:
The patent transforms the detection approach by shifting from relying on motion magnitude (1D optical flow vector length) to utilizing spatial distribution patterns across multiple dimensions. Image pyramids provide multi-scale spatial information, and saliency maps extract distributed feature patterns, enabling accurate detection of distant objects with small motion vectors by analyzing their spatial characteristics rather than motion magnitude alone.
Solution Approach 2:
The patent changes the detection parameters from raw optical flow vector values to processed features including saliency map values and image pyramid representations. By transforming the parameter space from direct motion measurement to multi-scale feature analysis, the system maintains detection accuracy across varying object distances where traditional optical flow parameters fail.
2Measurement precision
If image processing operations are performed to generate image feature maps, then detection accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the complex image processing task into distinct modular stages: generating image pyramids at multiple scales, computing saliency maps through localized operations, extracting pixel groups based on threshold criteria, and integrating results into final feature maps. This segmentation allows each stage to be optimized independently and enables parallel processing of different image regions and scales, reducing overall computational burden while maintaining accuracy.
Solution Approach 2:
The patent extracts only the most informative elements from the image data by identifying and processing salient pixels that exceed brightness thresholds. Instead of analyzing all pixels uniformly, the system extracts and processes only the subset of pixels that contribute most to object detection, significantly reducing computational complexity while preserving detection accuracy.
Data Source
AI summary
An image feature map generating unit (3) generates, on the basis of feature amounts extracted from a plurality of images successively captured by a camera (109), an image feature map which is an estimated distribution of the object likelihood on each of the images. An object detecting unit (4) detects an object on the basis of the image feature map generated by the image feature map generating unit (3).