3D Point Cloud Obstacle Recognition via 4D Array Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current obstacle detection systems for driverless vehicles using 2D vision and 3D sensing with laser radar have low accuracy, particularly in recognizing medium-sized and large-sized vehicles, due to high costs and immature implementation modes.
Innovation Solution
A method involving the conversion of 3D point cloud data into a four-dimensional array using different view angles and dimension data, which is then processed using a deep learning algorithm, such as a Convolution Neural Network, to improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If 2D vision technology is used for obstacle detection, then the cost is reduced, but the recognition accuracy deteriorates
Solution Approach 1:
The patent transforms 3D point cloud data into a 4D array by adding a channel dimension, mapping spatial information from three different view angles (front, side, top) into separate channels while preserving depth information. This dimensional transformation enables the use of成熟 deep learning algorithms designed for multi-channel data, achieving high-accuracy obstacle recognition without relying on expensive laser radar hardware
2Measurement precision
If 3D sensing technology with laser radar is used, then measurement precision is improved, but cost increases
Solution Approach 1:
The patent creates multiple virtual views (front, side, top perspectives) of the obstacle from a single 3D point cloud dataset. By generating these different perspective copies and mapping them to separate channels of a 4D array, the system achieves comprehensive spatial understanding without requiring multiple physical sensors or expensive laser radar equipment
Solution Approach 2:
The patent adds a channel dimension to transform 3D point cloud data into a 4D array structure, enabling the representation of multiple view angles and spatial features in a format suitable for deep learning processing. This dimensional expansion allows comprehensive obstacle characterization using computationally efficient methods
3Device complexity
If conventional 3D sensing algorithms are used, then implementation is simpler, but recognition accuracy deteriorates
Solution Approach 1:
The patent transforms 3D point cloud data into a 4D array by adding a channel dimension, mapping spatial information from three different view angles (front, side, top) into separate channels while preserving depth information. This dimensional transformation enables the use of成熟 deep learning algorithms designed for multi-channel data, achieving high-accuracy obstacle recognition
Solution Approach 2:
The patent applies normalization to the point cloud data before mapping it to the 4D array structure. This parameter transformation standardizes the spatial coordinates and intensity values, improving the effectiveness of subsequent deep learning processing and enhancing recognition accuracy across different lighting and distance conditions
4Speed
If simple obstacle detection is used, then processing speed is improved, but recognition accuracy deteriorates
Solution Approach 1:
The patent performs preliminary processing of the 3D point cloud data by organizing it into a structured 4D array format with normalized coordinates and multiple view angle channels before feeding it to the deep learning model. This pre-processing step optimizes the data structure for efficient processing, enabling the network to quickly extract features and achieve high-accuracy recognition
Data Source
AI summary
The present disclosure provides an obstacle type recognizing method and apparatus, a device and a storage medium, wherein the method comprises: obtaining 3D point cloud data corresponding to a to-be-recognized obstacle; mapping the 3D point cloud data and its dimension data to a four-dimensional array; recognizing a type of the obstacle through a deep learning algorithm based on the four-dimensional array. The solution of the present disclosure can be applied to determine the type of the obstacle such as a person, a bicycle or a motor vehicle; and recognize a small-sized vehicle, a medium-sized vehicle and a large-sized vehicle; and improve the accuracy of a recognition result.