Depth Image Object Detection for Reflective Surfaces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection systems inaccurately extract pixels from reflective surfaces, transparent objects, and changing backgrounds, leading to the inclusion of virtual images and unnecessary information, which complicates the identification of actual moving objects.
Innovation Solution
The system employs a method to differentiate between background and moving objects by determining the depth of pixels relative to the background, using a threshold value for depth differences and object size to accurately extract moving objects, and dynamically updates the background depth map to account for changing environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If background subtraction method is used to extract moving objects, then moving objects can be detected, but virtual images from reflective surfaces are also extracted as false positives
Solution Approach 1:
The patent segments the detection space into multiple depth regions by dividing the depth image into a first depth image (nearer side) and a second depth image (farther side) based on a reference depth value. This segmentation allows the system to process and compare depth information from different spatial regions separately, enabling distinction between real objects and virtual images caused by reflections.
Solution Approach 2:
The patent introduces depth as an additional dimension for object differentiation. By comparing depth values in the first depth image with depth values in the second depth image, the system can identify objects that exist only in one depth layer, thereby distinguishing real moving objects from virtual images that appear at different depth positions.
2Measurement precision
If depth difference method is used to detect objects behind transparent barriers, then all depth differences are detected, but unnecessary information from behind barriers is also extracted
Solution Approach 1:
The patent segments the detection space into multiple depth regions by dividing the depth image into a first depth image (nearer side) and a second depth image (farther side) based on a reference depth value. This segmentation allows the system to process and compare depth information from different spatial regions separately, enabling distinction between real objects and virtual images caused by reflections.
Solution Approach 2:
The patent applies different processing rules to different depth regions. For pixels in the first depth image (nearer side), the system extracts objects that do not have corresponding pixels in the second depth image, while for pixels in the second depth image, the system extracts objects that have corresponding pixels in the first depth image. This local quality approach ensures that only relevant objects within the desired detection range are extracted.
3Adaptability or versatility
If background depth map is updated dynamically, then changing backgrounds are adapted to, but false positives may increase from frequent updates
Solution Approach 1:
The patent implements dynamic background adaptation by updating the background depth map based on extracted moving objects. The system determines whether to update the background depth map by checking if the extracted moving object is a false positive, and only updates when appropriate. This dynamic approach allows the system to adapt to changing backgrounds while maintaining detection reliability through conditional update logic.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach ensures accurate extraction of moving objects, preventing false positives from reflective surfaces and transparent objects, and adapting to changing backgrounds, thereby improving the precision of object detection.
Implementation Method 1
TOF (a method of measuring the time until light is reflected by an object and returns) is used as a method of measuring a depth
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The present invention accurately extracts a moving object desired to be extracted regardless of the background condition of the moving object to be extracted. An object detection system (1) comprises a distance image detection means that detects a distance image from the outside world and a moving object extraction means (2) that extracts a moving object desired to be extracted from the distance image. The moving object extraction means records in advance a distance image in a state where the moving object desired to be extracted is not present in a storage unit (2a) as a background, and extracts only pixels for which the current distance is closer than the distance of the background, as candidates of pixels corresponding to the moving object to be extracted (S2, S3). In addition, the moving object extraction means: performs determination (S7) on the basis of the size of a cluster and so forth and extracts the moving object to be extracted from a pixel group composed of extracted candidate pixels; and newly records, as background distances, the distances of pixels included in parts not extracted as the moving object to be extracted, such recording performed in the storage unit (S8).