3D Position Detection Using Image-Guided Point Cloud Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Processing large-scale point cloud data is computationally costly, and existing methods that narrow down the point cloud range based on two-dimensional image recognition result in a large quadrangular pyramid, including unnecessary points from shielding objects, leading to decreased accuracy.
Innovation Solution
A position detection device and method that acquires three-dimensional point cloud information and multiple images from different viewpoints, uses image recognition to detect the target object's region, and specifies the corresponding point cloud region to accurately determine the target object's position in three-dimensional space, even with other objects in front or behind it.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the point cloud range is narrowed down to a quadrangular pyramid based on two-dimensional image recognition, then the processing speed is improved, but the accuracy decreases due to inclusion of unnecessary points from shielding objects
Solution Approach 1:
The patent segments the point cloud processing by dividing it into multiple regions based on depth information. It separates the foreground region (containing the target object) from the background region, processing each region with appropriate methods. This allows speed optimization for the foreground while maintaining accuracy by separately handling background points that would otherwise be incorrectly included as shielding objects.
Solution Approach 2:
The patent introduces depth information as an additional dimension to the traditional two-dimensional image recognition approach. By utilizing the depth axis, it creates a three-dimensional region of interest that accurately bounds the target object, preventing inclusion of background points while maintaining a compact processing volume for speed.
2Adaptability or versatility
If the quadrangular pyramid height is set from shortest to longest measurable distance, then all potential target positions are covered, but the point cloud range becomes very large requiring excessive computation
Solution Approach 1:
The patent performs preliminary action by using two-dimensional image recognition to detect the target object's position and dimensions before processing the point cloud. This preliminary detection provides prior information about the target's location and size, which is then used to define a precise three-dimensional region of interest, avoiding the need to process the entire large-scale point cloud.
Solution Approach 2:
The patent applies local quality by using different processing strategies for different regions of the point cloud. The foreground region containing the detected target is processed with high precision methods, while the background region is handled differently or excluded, optimizing computation based on the local characteristics of each region.
3Productivity
If image recognition is used to narrow down the point cloud range, then processing speed is improved, but shielding objects in front of the target are incorrectly identified as the target
Solution Approach 1:
The patent uses depth information as an intermediary to mediate between the two-dimensional image recognition results and the final target identification. The depth data provides additional verification, allowing the system to distinguish between foreground objects (true targets) and background objects (shielding objects) that may appear similar in the two-dimensional image alone.
Data Source
AI summary
There is provided a position detection device that recognizes a presence position of a target object in a three-dimensional space. The device acquires three-dimensional point cloud information of the space and a plurality of images obtained by imaging an area including surroundings of an object in the space from different imaging points. A region detection unit receives, as an input, the plurality of acquired images, determines whether a target object appears in the plurality of images, and detects a region of the object in each of the plurality of images in a case where the target object appears in each of the images. A specifying unit specifies a region of a point cloud corresponding to the target object based on the point cloud information and the region of the object detected in each of the images. A position detection unit specifies a position of the target object in the space by recognizing points corresponding to the target object from the point cloud information of the region specified by the specifying unit.