Stereo Camera Obstacle Detection via Semantic Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing obstacle-detection systems for vehicles emit electromagnetic or light waves, making vehicles detectable and limiting their use in stealth applications, while also being inefficient in detecting diverse types of obstacles accurately.
Innovation Solution
A method utilizing stereo cameras to acquire raw images, implementing semantic segmentation and disparity calculation algorithms to produce augmented images that provide 3D coordinates and obstacle identification, without emitting waves.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RADAR or LIDAR technology is used to detect obstacles, then obstacle detection capability is improved, but the vehicle becomes easily detectable by electromagnetic or light waves
Solution Approach 1:
The patent replaces electromagnetic wave-based detection systems (RADAR/LIDAR) with a passive optical system using stereo cameras. The system captures images from left and right cameras, processes them through rectification and disparity calculation to generate depth information and detect obstacles without emitting any electromagnetic or light waves, thereby maintaining stealth characteristics.
2Object-affected harmful factors
If passive obstacle detection system with small-field sensors is used, then vehicle stealth is maintained, but obstacle detection is limited to cable type only
Solution Approach 1:
The patent applies semantic segmentation to classify different types of obstacles (cables, pylons, trees, etc.) by analyzing image content. The system divides the scene into different obstacle categories through image processing algorithms, enabling detection of multiple obstacle types beyond just cables while maintaining passive sensor operation.
Solution Approach 2:
The patent transitions from 2D small-field sensor detection to 3D obstacle detection by implementing stereo vision with left and right cameras. Through disparity calculation and depth map generation, the system adds the depth dimension, enabling accurate detection and classification of various obstacle types in three-dimensional space while maintaining stealth.
3Reliability
If StereoVision with UV mapping is used to detect obstacles, then obstacle detection is achieved, but accuracy is insufficient and obstacle type differentiation is incapable
Solution Approach 1:
The patent replaces inaccurate UV mapping StereoVision with a refined stereo vision system that uses rectification to correct geometric distortions and disparity calculation to accurately compute depth. This substitution of the processing method significantly improves measurement precision for both obstacle detection and type differentiation.
Solution Approach 2:
The patent introduces intermediate processing steps (rectification and disparity calculation) as mediators between raw stereo images and final obstacle detection. The rectification corrects geometric distortions, and the disparity calculation generates accurate depth information, enabling precise obstacle detection and classification that UV mapping cannot achieve.
Data Source
AI summary
A method for detecting obstacles includes implementing a semantic segmentation algorithm to produce a first segmented image from a first raw left or right image; rectifying the raw stereo images to obtain rectified stereo images, and the first segmented image to obtain a first rectified segmented image; producing a disparity map; producing a second rectified segmented image; producing a list of predefined instances of obstacles present in the vehicle's environment from the rectified segmented stereo images; and integrating the predefined instances of obstacles into intermediate images obtained from the rectified stereo images.