LiDAR Decorrelation for Cluttered Autonomous Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection systems in autonomous vehicles face challenges in accurately detecting objects using LiDAR data alone, particularly in cluttered environments, as they often discard relevant information and suffer from biases due to the combination of closed world and open world object detection algorithms.
Innovation Solution
The implementation of a hybrid object detection system that combines closed world and open world object detection algorithms, using LiDAR data and camera images to correct errors and enhance detection accuracy by modifying point cloud segment sizes based on the number of data points from both algorithms, and employing a Camera-LiDAR Fusion (CLF) algorithm for improved segmentation and object detection in cluttered scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR data alone is used for object detection, then the detection speed is fast, but the detection accuracy deteriorates in cluttered environments where objects are in close proximity
Solution Approach 1:
The patent combines LiDAR data with camera image data to perform object detection. The LiDAR point cloud data is projected onto the camera image plane, and objects are detected by analyzing both the depth information from LiDAR and the visual information from the camera. This fusion approach improves detection accuracy in cluttered environments while maintaining reasonable processing speeds.
2Measurement precision
If LiDAR and camera data are combined for object detection, then the detection accuracy is improved, but the processing complexity increases
Solution Approach 1:
The patent extracts only the relevant portions of LiDAR data by projecting the point cloud onto the camera image plane. Instead of processing all LiDAR points, only those that map to the image region containing objects of interest are processed. This extraction approach reduces processing complexity while maintaining the benefits of multi-sensor fusion.
3Measurement precision
If traditional segmentation algorithms are used, then the processing is simple, but the segmentation accuracy deteriorates when objects are in close proximity
Solution Approach 1:
The patent introduces a new dimension for segmentation by projecting 3D LiDAR points onto the 2D camera image plane. Objects are segmented by analyzing their projections in the image space, which provides additional spatial context. This dimensional transformation enables more accurate separation of closely spaced objects compared to traditional 3D point cloud segmentation methods.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach improves the accuracy and completeness of object detection, reduces end-to-end tracking latency, and enhances the ability to detect objects in close proximity, while avoiding biases and double counting of sensor data, leading to more reliable autonomous driving operations.
Implementation Method 1
The LiDAR detectors generate LiDAR datasets that measure the distance from the vehicle to an object at a plurality of different times
Data Source
AI summary
Systems/methods for operating an autonomous vehicle. The methods comprise: obtaining LiDAR data generated by a LiDAR system of the autonomous vehicle and image data generated by a camera of the autonomous vehicle; performing a first object detection algorithm to generate first object detection information using the LiDAR data, and a second object detection algorithm to generate second object detection information using both the LiDAR data and the image data; and processing the first and second object detection information to determine whether a given object was detected by both the first and second object detection algorithms. When a determination is made that the given object was detected by both the first and second object detection algorithms, the first object detection information is selectively modified based on contents of the second object detection information. The modified first object detection information is used to facilitate at least one autonomous driving operation.