Sparse 3D Convolutional Network for Point Cloud Target Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The rapid development of electric vehicles requires efficient and accurate three-dimensional target detection methods for driver assistance systems, particularly in handling large amounts of point cloud data with significant interference, to enhance processing speed, generalization, and anti-interference capabilities.
Innovation Solution
A target detection method involving voxelization of point cloud data, followed by feature extraction using a sparse 3D convolutional network and a detection head network, which includes coordinate transformation, normalization, and binary classification for yaw angle determination, to improve detection accuracy and range without increasing computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional point cloud processing methods are used, then detection accuracy can be maintained, but processing speed is slow due to huge data amount
Solution Approach 1:
The patent segments the dense point cloud data into multiple sparse point clouds based on detection boxes from a two-stage detector. This segmentation divides the huge data amount into smaller, manageable portions that can be processed faster while maintaining detection accuracy for each segment.
Solution Approach 2:
The patent transforms the processing approach by introducing dimensionality reduction through sparse point cloud representation. By converting dense 3D point clouds into sparse representations based on detection regions, the method reduces computational dimensions and accelerates processing speed.
2Reliability
If dense point cloud data is processed, then detection accuracy is maintained, but anti-interference capability is reduced due to large amount of interference information
Solution Approach 1:
The patent extracts only the relevant sparse point clouds within detection boxes identified by the two-stage detector, removing unnecessary interference information from the surrounding areas. This extraction process maintains detection accuracy while significantly reducing anti-interference challenges.
Solution Approach 2:
The patent applies local quality processing by focusing computational resources on specific regions of interest (detection boxes) rather than uniformly processing the entire point cloud. This localized approach improves anti-interference capability by concentrating analysis where targets are likely to be found.
3Measurement precision
If complex feature extraction networks are used, then detection accuracy is improved, but processing speed decreases due to increased computational load
Solution Approach 1:
The patent segments the feature extraction process into two stages: a coarse two-stage detector for initial target identification and a refined sparse point cloud processor for accurate detection. This segmentation allows complex feature extraction to be applied only where necessary, maintaining accuracy while improving overall processing speed.
Solution Approach 2:
The patent applies partial action by using simplified processing for most of the point cloud and reserving complex feature extraction only for regions containing potential targets. This selective application of computational complexity maintains detection accuracy while significantly reducing overall processing time.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The disclosure relates to the field of electric vehicles, and specifically provides a target detection method, a storage medium, an electronic device, and a vehicle, to solve a problem of how to improve a processing speed, generalization, and an anti-interference capability of point cloud data to implement rapid and accurate three-dimensional target detection. For this purpose, the method of the disclosure includes: voxelizing point cloud data to be processed, and obtaining a target detection result by using a target detection network including a sparse 3D convolutional network, a two-dimensional detection network, and a detection head network in sequence. According to the method of the disclosure, point cloud data is voxelized based on voxel positions, thereby reducing dependence of the point cloud data on an absolute distance, and improving anti-interference and generalization of the network. A downsampling structure of the sparse 3D convolutional network is optimized, thereby expanding a detection range in a height direction without an increase in a calculation amount. A first orientation attribute and a second orientation attribute of a yaw angle are introduced, thereby improving a yaw angle generation method, and improving yaw angle detection accuracy.