Lidar Point Cloud Feature Map for Real-Time Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection methods in autonomous driving systems, such as those using cameras and radar sensors, face challenges in accuracy and precision, particularly in determining depth and object type, limiting their effectiveness for real-time object detection.
Innovation Solution
A method and device for real-time object detection based on lidar point clouds, which generates a feature map through deep learning and rule-based processing, combining deep-learning-based and rule-based feature extraction to enhance object detection performance and speed by projecting 3D points into 2D bird's eye views and recording feature information in lattices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If camera-based object detection is used, then 2D detection is easy, but depth determination becomes difficult
Solution Approach 1:
The patent combines camera-based 2D detection with lidar-based 3D point cloud data to merge the advantages of both systems. The camera provides easy 2D image analysis while the lidar provides accurate depth information through its point cloud data, resolving the contradiction between detection ease and depth accuracy.
Solution Approach 2:
The patent uses a feature map as an intermediary that integrates both camera images and lidar point cloud data. This feature map serves as a mediator that combines 2D visual information with 3D spatial information, enabling both easy 2D detection and accurate depth determination simultaneously.
2Reliability
If radar sensor is used, then weather resistance and distance resolution are good, but object detection precision deteriorates
Solution Approach 1:
The patent merges radar sensor data with lidar point cloud data and camera images. The radar provides weather resistance and distance resolution, while the lidar and camera provide object type identification and detection precision, achieving both reliability and precision through combination.
Solution Approach 2:
The patent creates a composite detection system that integrates multiple sensor types (radar, lidar, camera) into a unified feature map. This composite approach allows the system to leverage the weather resistance of radar while compensating for its poor object detection precision through lidar and camera data.
3Measurement precision
If deep learning-based feature extraction is used, then detection performance improves, but computational complexity increases
Solution Approach 1:
The patent segments the feature extraction process into multiple stages: first extracting features from individual sensor data (camera images and lidar point clouds) separately, then combining these features into a unified feature map. This segmentation reduces computational complexity by processing data in manageable stages rather than simultaneously.
Solution Approach 2:
The patent performs preliminary feature extraction from camera and lidar data before combining them into the final feature map. This preliminary action allows the system to pre-process and simplify the data structure before the main detection task, reducing overall computational complexity while maintaining detection performance.
4Measurement precision
If 3D point cloud data is processed directly, then detection accuracy improves, but processing speed decreases
Solution Approach 1:
The patent transforms the 3D point cloud data into a 2D feature map representation. This dimensionality change allows the system to maintain the spatial accuracy information from 3D data while processing it in a 2D format that is computationally faster, thus improving processing speed without sacrificing detection accuracy.
Solution Approach 2:
The patent creates a feature map that is essentially a processed copy or representation of the original 3D point cloud data. This feature map copy retains the essential spatial and detection information while being optimized for faster processing, allowing the system to work with a simplified representation rather than the full 3D data.
Data Source
AI summary
Provided are a method, computing device, and recording medium for detecting an object in real time on the basis of a lidar point cloud. The method is performed by the computing device and includes generating a feature map on the basis of a point cloud which is generated by scanning a certain region through a lidar sensor and deriving an object detection result by inputting the generated feature map to a pretrained object detection model.