Pedestrian Point Cloud Detection Using Stable Region Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In autonomous driving scenarios, accurately detecting pedestrian movement is challenging due to their unpredictable posture changes, which can affect the safety and reliability of autonomous vehicles.
Innovation Solution
A method and apparatus that project point cloud data of pedestrians onto a ground coordinate system to determine the direction of the connection line between their shoulders, extract a stable region with minimal morphological changes, and analyze the center point coordinates across consecutive frames to determine movement information, reducing the influence of posture changes on detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the entire point cloud data of the pedestrian is used for movement detection, then the detection coverage is comprehensive, but the detection accuracy is reduced due to posture changes affecting multiple regions
Solution Approach 1:
The patent divides the pedestrian point cloud data into multiple regions based on morphological stability characteristics. By segmenting the pedestrian model into stable regions (head, torso) and unstable regions (limbs), the system can selectively use only the stable regions for movement detection, thereby improving accuracy while reducing the complexity of processing entire point cloud data.
Solution Approach 2:
The patent applies different processing strategies to different regions of the pedestrian model. Stable regions are identified and weighted more heavily for movement detection, while unstable regions are either excluded or given lower weight. This local differentiation allows the system to optimize detection accuracy by focusing on regions with minimal morphological changes.
2Measurement precision
If traditional point cloud methods are used without region extraction, then the detection process is simple, but the detection accuracy is low due to strong walking arbitrariness of pedestrians
Solution Approach 1:
The patent performs preliminary extraction of stable regions from the pedestrian point cloud data before movement detection. By pre-identifying and isolating stable regions (head, torso) that exhibit minimal morphological changes during walking, the system prepares optimized data structures in advance, reducing computation time during actual detection while improving accuracy.
Solution Approach 2:
The patent changes the selection criterion from using all point cloud data to selecting only points within stable regions based on morphological stability parameters. This parameter change filters out noisy data from unstable regions, improving detection accuracy while the efficient region extraction algorithm minimizes the time loss.
3Reliability
If all regions of the pedestrian are monitored for movement, then the detection coverage is complete, but the reliability is reduced due to morphological changes in unstable regions
Solution Approach 1:
The patent segments the pedestrian detection area into stable and unstable regions, then selectively monitors only the stable regions for movement detection. This segmentation improves reliability by excluding regions with strong morphological changes (limbs, head) while maintaining sufficient coverage through the stable torso and head regions, which are sufficient for detecting overall pedestrian movement.
Solution Approach 2:
The patent extracts and isolates the stable region point cloud data from the complete pedestrian point cloud. By taking out only the reliable stable regions (particularly the torso and head) for movement analysis, the system improves detection reliability while the extraction process is designed to be computationally efficient, minimizing time loss.
Data Source
AI summary
A method, including: acquiring a point cloud frame including point cloud data of a pedestrian; projecting the point cloud data of the pedestrian to a ground coordinate system to obtain projection point data of the pedestrian; determining a direction of a connection line between the two shoulders of the pedestrian, based on a location distribution of the projection point data; extracting a point cloud of a stable region from the point cloud data of the pedestrian based on the direction of the connection line between the two shoulders of the pedestrian, a form change range of the stable region when the pedestrian moves being smaller than form change ranges of other regions of the pedestrian; and determining movement information of the pedestrian based on a coordinate of a center point of the point cloud of the stable region in a plurality of consecutive point cloud frames.