Monocular Object Detection Using Map-Derived Depth Layers
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
A monocular-based object detection system that utilizes a road map to generate additional layers of information, such as ground height, depth, and drivable area features, which are superimposed onto images captured by a vehicle's camera, enabling object detection and trajectory prediction through machine-learning algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR-based object detection is used, then object detection accuracy is improved, but system cost increases and weather sensitivity worsens
Solution Approach 1:
The patent creates a synthetic depth map that copies and simulates the depth information normally provided by LiDAR, but generates it computationally from monocular camera images and map data. This synthetic depth map serves as a substitute for actual LiDAR measurements, achieving similar object detection functionality without the associated cost and weather sensitivity issues
Solution Approach 2:
The patent introduces map information as an intermediary element that bridges the gap between 2D monocular images and 3D spatial understanding. By overlaying map data (road geometry, elevation, drivable areas) onto the monocular image, the system creates additional contextual layers that enable depth estimation and object detection without requiring direct depth sensing hardware
2Measurement precision
If LiDAR-based object detection is used, then object detection accuracy is improved, but system cost increases
Solution Approach 1:
The patent replaces expensive LiDAR hardware with a combination of inexpensive monocular cameras and computational algorithms. The system uses readily available, low-cost components (standard camera, map data) to achieve object detection functionality that would otherwise require costly specialized hardware
Solution Approach 2:
The patent creates a synthetic depth map that copies and simulates the depth information normally provided by LiDAR, but generates it computationally from monocular camera images and map data. This synthetic depth map serves as a substitute for actual LiDAR measurements, achieving similar object detection functionality without the associated cost and weather sensitivity issues
3Object-affected harmful factors
If monocular-based object detection with map overlay is used, then system cost is reduced and weather resistance is improved, but detection precision may be affected
Solution Approach 1:
The patent transforms 2D monocular image data into 3D spatial understanding by integrating map information and generating synthetic depth maps. This dimensional transformation allows the system to infer depth, elevation, and spatial relationships from purely 2D inputs, compensating for the lack of direct depth sensing while maintaining detection precision
Solution Approach 2:
The patent performs preliminary processing of map data to pre-compute road geometry, elevation profiles, and drivable area boundaries before they are needed for object detection. This pre-processing creates ready-to-use contextual information that can be quickly integrated with real-time camera feeds, improving both speed and accuracy of detection
Data Source
AI summary
Disclosed herein are systems, methods, and computer program products for object detection. The methods comprise performing the following operations by a computing device: obtaining an image that comprises a plurality of layers superimposed on each other; identifying a center point of a robot on a map; selecting a portion of the map contained in a geometric shape overlaid on the map so as to have a center set to the center point of the robot; obtaining map information associated with the selected portion of the map; generating at least one additional layer using the map information; superimposing the at least one additional layer onto the image to generate a modified image; and performing an object detection algorithm to detect at least one object in the modified image.