Depth Image Object Detection via Floor Plane Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image segmentation technologies, such as those using Kinect sensors, face challenges in accurately detecting objects, especially when they contact a floor, due to noise from depth sensors and non-uniform backgrounds, leading to difficulties in separating the object from the background.
Innovation Solution
A system and method that utilize a depth image processing system with a communication module, memory, and processor to extract object areas from a depth image by setting predetermined floor and outer planes, calculating boundary values, and segmenting the image to distinguish between the object and the floor, thereby accurately extracting foreground pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If depth sensor is used to extract object area, then object detection can be performed in real time, but noise from depth sensor causes difficulty in separating object from background
Solution Approach 1:
The patent divides the depth image processing into multiple stages: first extracting a rough object area, then identifying a target area at boundaries, and finally extracting precise object areas. This multi-stage segmentation approach allows real-time processing while improving separation accuracy by focusing computational effort on critical boundary regions.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. High-precision boundary detection algorithms are applied specifically to target areas at object boundaries, while simpler methods are used for interior regions. This local quality approach maintains real-time performance while improving overall separation accuracy.
2Ease of operation
If learner extraction method is used, then person area can be easily extracted, but satisfactory performance is not obtained in outer portion
Solution Approach 1:
The patent segments the extraction process into two phases: a preliminary learner-based extraction for simplicity, followed by a boundary refinement stage that specifically processes target areas to improve outer portion detection accuracy.
Solution Approach 2:
The patent performs preliminary object area extraction using the learner method to establish a rough boundary, then uses this preliminary result to guide subsequent boundary refinement processing. This preliminary action enables the system to maintain ease of operation while setting up conditions for improved boundary detection.
3Device complexity
If depth value similarity is used to identify object and background, then processing is simplified, but object and background cannot be separated when they have similar depth values
Solution Approach 1:
The patent applies depth value similarity analysis selectively only to target areas at boundaries, not to the entire image. This local quality approach maintains low processing complexity for most regions while applying more sophisticated separation methods only where needed for accurate object-background separation.
Solution Approach 2:
The patent performs partial processing by focusing computational resources on boundary regions rather than analyzing every pixel in the image. This partial action approach keeps overall processing complexity low while achieving sufficient separation accuracy at critical boundaries.
Data Source
AI summary
Provided is a system for detecting an object from a depth image. The system includes a communication module, a memory, and a processor. By executing the object detection program, the processor extracts a first object area and a second object area from the depth image, based on a predetermined floor plane and an outer plane which is set with respect to the predetermined floor plane. The processor extracts a target area including pixels of the second object area which are spaced apart from the first object area by a predetermined interval. The processor samples a pixel, which is not included in the target area, to extract a floor area from the second object area, calculates a boundary value of an object and a floor, based on the floor area and the target area, and extracts a foreground pixel from the target area, based on the calculated boundary value.