Object Detection Device Using Multi-Height Candidate Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques based on time-series images from moving vehicles incur a high load of arithmetic processing, which is inefficient and resource-intensive.
Innovation Solution
An object detection device comprising an image input unit, movement information acquisition unit, feature point extraction unit, candidate point calculation unit, correspondence point detection unit, and object detection unit, which calculates candidate points at multiple heights based on movement information to reduce arithmetic processing load and efficiently detect objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing object detection techniques based on time-series images are used, then objects can be detected from moving vehicle images, but the arithmetic processing load becomes excessively high
Solution Approach 1:
The patent segments the image processing task by extracting only feature points (edges, corners, contours) from images rather than processing all pixels. This segmentation of processing targets dramatically reduces computational complexity while maintaining detection accuracy, as feature points contain the essential structural information needed for object detection.
Solution Approach 2:
The patent extracts only the necessary components (feature points and their correspondence relationships) from the complete image data. By taking out and processing only the critical feature points rather than entire images, the system achieves efficient object detection with significantly reduced arithmetic processing requirements.
2Reliability
If feature point correspondence is calculated between time-series images, then object detection can be performed, but the calculation complexity increases significantly
Solution Approach 1:
The patent applies local quality by focusing computational resources only on feature point locations rather than uniform processing of entire images. Each feature point is processed individually for correspondence detection, allowing the system to concentrate calculations where they are most needed while ignoring redundant areas, thus reducing overall complexity.
Solution Approach 2:
The patent replaces complex mechanical image matching processes with feature point-based correspondence detection. Instead of comparing entire image regions or performing exhaustive pixel-level matching, the system uses characteristic feature points as proxies, substituting a simplified computational model that achieves the same detection goal with much lower complexity.
3Measurement precision
If multiple candidate points are calculated at different heights, then object height detection becomes possible, but processing time increases
Solution Approach 1:
The patent performs preliminary action by pre-calculating candidate points at multiple heights based on the detected feature point correspondence. Rather than performing time-consuming height calculations after detection, the system prepares height candidate information in advance during the feature matching phase, enabling efficient object height determination without adding significant processing time.
Data Source
AI summary
An object detection device acquires a plurality of images captured by a camera mounted in a moving body and acquires movement information of the camera. The object detection device extracts an edge from a first image among the acquired images. The object detection device calculates a plurality of candidate points corresponding to the edge in a second image, based on the movement information and calculates candidate points corresponding to a plurality of heights as height candidates of the edge. The object detection device calculates a correlation index value between the edge and each of the plurality of candidate points and detects the candidate point with the largest correlation as a correspondence point of the edge. The object detection device detects an object at a position of the edge when the correlation index value of the detected correspondence point is equal to or smaller than a predetermined threshold.