Monocular Object Detection Using Road Map Layers Instead of LiDAR
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
LiDAR-based object detection systems are costly and sensitive to weather conditions, limiting their effectiveness in modern vehicles.
Innovation Solution
Implementing a monocular-based object detection system that uses information from a road map, including ground height, depth, drivable areas, and lane information, to generate a modified image for object detection, leveraging machine-learning algorithms to estimate object position, orientation, and spatial extent.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR based object detection is used, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent uses a monocular camera to capture images and creates a modified image by copying and overlaying road map information (ground height, depth, drivable areas, lane markings) onto the original image. This synthetic copy provides additional depth and spatial context without requiring complex LiDAR hardware, thereby achieving improved measurement precision while reducing device complexity
Solution Approach 2:
The patent introduces road map information as an intermediary element that mediates between the simple monocular camera input and the complex spatial understanding required for accurate object detection. By overlaying preprocessed map data (ground planes, drivable areas, lane directions) onto the image, the system bridges the gap between 2D image data and 3D spatial awareness without direct use of complex sensors
2Reliability
If LiDAR based object detection is used, then object detection capability is improved, but sensitivity to weather conditions worsens
Solution Approach 1:
The system creates a modified image by copying road map information onto the original image captured by the monocular camera. This approach uses optical information that is inherently more robust to weather conditions compared to LiDAR, as cameras continue to function in rain, fog, and snow where LiDAR performance degrades. The copied map data provides consistent spatial references regardless of weather
Solution Approach 2:
The patent replaces the mechanical/optical LiDAR system with a camera-based optical system. Cameras are generally more weather-resistant than LiDAR sensors, which rely on laser propagation that is easily scattered by precipitation and fog. The substitution of LiDAR's active optical sensing with passive optical sensing via camera improves reliability under adverse weather conditions
3Ease of manufacture
If monocular based object detection with road map information is used, then device cost is reduced, but measurement precision may worsen
Solution Approach 1:
The patent performs preliminary processing of road map information before it is needed for object detection. Ground height maps, depth maps, drivable area masks, and lane direction information are precomputed and stored. During actual operation, these preprocessed data structures are simply overlaid onto the camera image, avoiding the need for complex real-time computations and enabling accurate object detection with lower-cost hardware
Solution Approach 2:
The patent enhances 2D monocular image data by adding multiple information layers from road maps (ground height as vertical dimension, depth distance, drivable area boundaries, lane directions). This multi-dimensional enrichment of the image data compensates for the inherent limitations of monocular vision, providing pseudo-3D spatial context that improves measurement precision without requiring expensive depth-sensing hardware
4Loss of information
If multiple additional layers are superimposed on the image, then information completeness is improved, but device complexity increases
Solution Approach 1:
The patent segments road map information into distinct functional layers: ground height layer, ground depth layer, drivable geographical area layer, map point distance-to-lane center layer, lane direction layer, and intersection layer. Each layer is independently generated and then overlaid onto the original image. This segmentation allows the system to process and integrate multiple types of spatial information in a modular fashion, improving information completeness while managing processing complexity through structured organization
Data Source
AI summary
Systems and methods for object detection. The methods comprise: obtaining, by a computing device, an image comprising a plurality of color layers superimposed on each other; generating at least one first additional layer using information contained in a road map (where the first additional layer includes ground height information, ground depth information, drivable geographical area information, map point distance-to-lane center information, lane direction information, or intersection information); generating a modified image by superimposing the first additional layer on the color layers; and causing, by the computing device, control of a vehicle's operation based on the object detection made using the modified image.