3D Object Detection Using Virtual Cuboid Density Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D object detection systems using a single camera face challenges in accurately detecting 3D objects due to variations in background and lighting conditions, leading to false positive and false negative detections, which reduces position accuracy and reliability, especially in traffic impediment warning systems.
Innovation Solution
A 3D object detecting apparatus and method that uses a single image-capture device to create a detection image by eliminating unnecessary regions, and generates a density map by projecting virtual cuboids onto the image-capture surface, allowing for high-density region extraction and accurate position detection of 3D objects, even in complex scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single camera is used for 3D object detection, then device complexity is reduced, but detection accuracy deteriorates due to false positive and false negative detections
Solution Approach 1:
The patent segments the image-capture surface into multiple sub-regions and divides the 3D object detection task into multiple independent detection processes, one for each sub-region. This segmentation allows the system to handle complex scenes more effectively by processing smaller, more manageable portions of the image, thereby reducing false detections while maintaining the simplicity of using a single camera.
Solution Approach 2:
The patent introduces a vertical dimension by detecting not only the horizontal position but also the vertical position (height) of 3D objects relative to the camera. By adding this vertical dimension to the detection process and creating a density map that includes height information, the system achieves more accurate object localization without requiring multiple cameras, thus resolving the contradiction between device simplicity and detection accuracy.
2Measurement precision
If complex processing such as pattern matching is performed to correct false detections, then detection accuracy is improved, but processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary actions by pre-dividing the image-capture surface into sub-regions and pre-establishing the detection framework before actual object detection occurs. This preliminary preparation allows the system to efficiently process images without requiring time-consuming pattern matching during the detection phase, as the structure and guidance for accurate detection are already in place.
Solution Approach 2:
The patent creates a density map that copies and accumulates detection information from multiple sub-region detections into a unified representation. This density map serves as a compressed summary of the detection results, allowing the system to quickly identify accurate object positions without reprocessing the entire image through complex pattern matching algorithms.
3Measurement precision
If background regions are eliminated to extract only 3D objects, then detection precision is improved, but false positive detections still occur due to variations in lighting and background conditions
Solution Approach 1:
The patent applies local quality by treating different sub-regions of the image-capture surface with specialized detection characteristics. Each sub-region is processed with detection parameters optimized for its specific location and content, allowing the system to adapt to local variations in lighting and background conditions. This localized approach enables more reliable 3D object detection while minimizing false positives caused by uniform background elimination methods.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A 3D-object detecting apparatus (1) includes a detection-image creating device (4) that detects a 3D object on an image-capture surface from an image captured by an image-capture device (2) and that creates a detection image in which a silhouette of only the 3D object left; a density-map creating device (5) that determines the 3D object's spatial densities at corresponding coordinate points in a coordinate plane on the basis of the detection image and mask images obtained for the corresponding coordinate points on the basis of virtual cuboids arranged for the corresponding coordinate points and that creates a density map having pixels for the corresponding coordinate points such that the pixels have pixel values corresponding to the determined spatial densities; and a 3D-object position detecting device (6) that detects the position of the 3D object as a representative point in a high-density region in the density map.