Stereo Camera Depth Map Drivable Space Boundary Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current driver aids, such as those using ultrasound, are limited in detecting all objects due to reliance on strong signal reflections, making them unreliable for identifying objects at high speeds or in various environments, and are not suitable for autonomous vehicles that need to plot safe paths in diverse conditions.
Innovation Solution
A method using a calibrated stereo camera to create a depth map and range bin histograms, identifying boundary points to determine the drivable space by analyzing pixel counts and thresholds, without requiring training data or machine learning, allowing detection of objects above a 3D plane and presenting the boundary on a 2D image for visual representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ultrasound sensors are used to detect objects, then the system is low cost and reliable for close objects, but it cannot detect all objects at high speeds or in various environments
Solution Approach 1:
The patent replaces mechanical contact-based sensing (ultrasound reflection requiring physical proximity) with optical field-based sensing (stereo vision). The stereo camera system uses optical triangulation to calculate depth and detect objects without requiring strong acoustic reflections, enabling detection across diverse environments and speeds while maintaining reliability.
Solution Approach 2:
The stereo camera system serves multiple functions: it detects objects at various distances, operates at all speeds, works in diverse lighting conditions, and provides both 2D and 3D spatial information. This multi-functional capability replaces the limited function of ultrasound sensors which only work reliably for close objects requiring strong reflections.
2Adaptability or versatility
If stereo cameras are used to create depth maps, then detection versatility improves, but processing complexity increases
Solution Approach 1:
The patent segments the complex depth map processing into distinct stages: first creating a depth map from stereo pairs, then generating range bin histograms that divide space into discrete depth intervals, and finally analyzing pixel counts in each bin. This segmentation transforms a complex continuous 3D processing problem into manageable discrete steps, reducing overall computational complexity while maintaining detection versatility.
Solution Approach 2:
The patent processes only the necessary portions of the depth map by using range bin histograms that focus on specific depth intervals relevant to drivable space detection. Rather than analyzing every pixel in the entire depth map, the system selectively processes regions within relevant range bins, reducing computational load while maintaining detection accuracy for critical areas.
3Measurement precision
If range bin histograms are used to identify boundaries, then detection accuracy improves, but computational requirements increase
Solution Approach 1:
The patent creates a simplified 1D histogram representation (copy) of the 3D depth map data. Instead of directly analyzing the complex 3D spatial relationships in the depth map, the system copies the essential depth distribution information into range bin histograms, which can be processed more efficiently. This copying approach maintains boundary detection accuracy while significantly reducing computational requirements.
Solution Approach 2:
The patent transforms the depth map data from continuous 3D spatial coordinates into discrete range bin counts. By changing the parameter representation from continuous depth values to discrete histogram bin counts, the system enables more efficient computational processing while maintaining the precision needed to identify boundary points where pixel counts exceed thresholds.
Data Source
Figure 1
Figure 2
Figure 3(a)~3(b)
AI summary
A method of determining the characteristics of a scene around a vehicle comprises capturing a stereo pair of images of the scene 200 and processing the images 210 to produce a depth map of the scene 220. Each pixel in the depth map is assigned a value that corresponds to the range of a corresponding region in the scene, the pixels being arranged in a grid of rows and columns with each column of pixels in the grid corresponding to a vertically oriented set of regions in the scene and each row a horizontally oriented set of regions in the scene. The values for one or more columns of pixels in the depth map are binned to form a corresponding histogram 230, each bin in the histogram having a count value that corresponds to the number of pixels in the column that have a depth within the range assigned to the bin. The or each of the range bin histograms are processed to determine for the or each histogram the lowest range bin that is indicative that an object that represents a non-drivable region is present at a depth that lies in the range of depths assigned to the bin, thereby identifying the location of one or more boundary points 280 that lie on a boundary of a drivable space in the scene.