Image Processing Device Height Layer Object Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection systems struggle to accurately separate objects with complex edges in gray images and detect obstacles near the road surface, especially when non-road surface areas occupy a large portion of the image and the road surface has a significant gradient, leading to incorrect grouping and detection of adjacent objects as a single region.
Innovation Solution
The mobile body control system generates multiple real U-maps in a layer structure based on the height of objects from the road surface, using disparity voting to create layer U-maps that integrate and separate grouping results, allowing for accurate detection of objects by efficiently utilizing height information and improving the separation of connected objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If objects are grouped using depth map and divided based on outline portions in gray images, then object detection can be performed, but adjacent objects with complex edges are incorrectly detected as a single region
Solution Approach 1:
The patent introduces height information from the depth map as an additional dimension to separate objects that are merged in 2D gray images. By creating multiple grouping results at different height layers and comparing them, the system can distinguish adjacent objects that have different heights, thereby resolving the contradiction between detection coverage and separation accuracy.
Solution Approach 2:
The patent segments the object detection process into multiple height layers by creating grouping results at different height thresholds. Each layer independently groups objects, and the final result is obtained by integrating these segmented layer results, which improves the separation of adjacent objects with complex edges.
2Area of stationary object
If object detection is performed in regions where objects having heights equal to or higher than the road surface can be present, then detection coverage is improved, but adjacent objects are detected in a connected manner
Solution Approach 1:
The patent uses the height dimension from depth information to separate objects that are connected in the 2D image plane. By generating multiple grouping results at different height layers and comparing their intersections, the system can identify and separate adjacent objects even when they occupy the same horizontal region, thus maintaining detection coverage while improving separation precision.
Solution Approach 2:
The patent changes the detection parameter from 2D spatial coordinates to 3D coordinates including height. By varying the height layer threshold and creating multiple grouping results at different height levels, the system can distinguish between objects that appear connected in 2D but are separated in 3D space, resolving the contradiction between coverage and precision.
3Measurement precision
If multiple grouping results at different height layers are generated and integrated, then separation of connected objects is improved, but processing complexity increases
Solution Approach 1:
The patent segments the complex processing task into multiple independent height layers, where each layer performs grouping independently. This segmentation allows parallel processing of different height levels and simplifies the overall complexity by breaking down the 3D grouping problem into multiple 2D grouping problems that can be solved using existing algorithms.
Solution Approach 2:
The patent manages processing complexity by organizing the multi-layer grouping results in a structured manner based on height layers. Each layer's grouping result is stored and compared systematically, allowing the system to handle the increased dimensionality without exponential complexity growth through efficient data organization and comparison strategies.
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 enables accurate and high-speed detection of objects, particularly persons, by correctly identifying and separating objects even when they are adjacent or have complex edges, enhancing the reliability of object detection and vehicle control systems.
Implementation Method 1
generates multiple real U-maps in a layer structure based on the height of objects from the road surface, using disparity voting to create layer U-maps that integrate and separate grouping results
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
An image processing device (3) includes: a reference detecting unit (46) configured to detect a reference object serving as a reference for a height of an object from a depth map including distance information on pixels; a map generating unit (49) configured to cast, out of a plurality of maps indicating respective height ranges as a height range from the reference object, votes from the distance information on the pixels on the maps indicating the height ranges corresponding to heights of the respective pixels in the depth map from the reference object; an aggregation detecting unit (50) configured to detect an aggregated region of the distance information from the maps; and a classifying unit (51) configured to classify a type of the object of the aggregated region based on distribution of the aggregated region on the maps.