3D Point Cloud Segmentation for Drone Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 3D scanning technologies, particularly LIDAR systems used in drones, face challenges such as high computational intensity for error correction, reduced accuracy due to inclusion of error points, and limited flight duration for high-resolution mapping, along with issues like less dense point data at the edge of the scan area and high costs of existing LIDAR technologies.
Innovation Solution
The method involves dividing a 3D point cloud into boxes, processing each box to detect flat planes, and merging adjacent boxes with shared data points to form objects, incorporating error correction into the detection process, which reduces computational intensity and improves accuracy, using a 'bottom-up' approach and potentially neural networking for real-time object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If mathematical functions are mapped to surfaces for object detection, then object detection can be performed, but computational intensity increases and accuracy decreases due to error points being included in the mapping function
Solution Approach 1:
The patent divides the 3D point cloud into multiple boxes or regions, processing each box independently to detect planes. This segmentation reduces the computational complexity by breaking down the large point cloud into smaller, manageable units, allowing error correction to be performed locally without affecting the entire dataset.
Solution Approach 2:
The patent extracts and removes error points from the point cloud data through statistical analysis and variance calculation. By identifying and removing outliers before plane detection, the system improves measurement precision without requiring computationally intensive error correction processes.
2Measurement precision
If error correction is performed on mapped surfaces, then measurement accuracy improves, but computational intensity increases and processing time increases
Solution Approach 1:
The patent performs preliminary error correction by removing error points from the point cloud before conducting plane detection. This preliminary action ensures that only clean data is used for object detection, improving accuracy while reducing the computational burden during the main detection process.
Solution Approach 2:
By segmenting the point cloud into boxes and performing error correction within each box rather than across the entire dataset, the system reduces processing time while maintaining correction accuracy. The localized approach allows parallel processing of multiple boxes simultaneously.
3Measurement precision
If high-resolution mapping is achieved, then measurement precision improves, but flight duration decreases due to energy consumption
Solution Approach 1:
The patent changes the parameters of the LIDAR scanning system, including scan density, resolution, and processing thresholds, to achieve high-resolution mapping while optimizing energy consumption. By adjusting these parameters, the system maintains mapping quality while extending operational flight duration.
4Area of stationary object
If LIDAR scanning coverage is increased, then field of view improves, but point data density decreases at the edges of the scan area
Solution Approach 1:
The patent applies different processing quality levels to different regions of the point cloud. Edge regions are processed with adjusted parameters to maintain adequate density, while central regions use standard processing. This local quality adjustment ensures consistent measurement precision across the entire field of view without requiring uniform high-density scanning everywhere.
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 results in faster and more accurate object detection in 3D space, reducing computational workload and enhancing the accuracy of object detection lists, while also being more cost-efficient and deployable in various applications.
Implementation Method 1
The scanning devices may be stationary or mounted on a mobile platform as needed to collect data from the desired field of view (FOV).
Data Source
AI summary
A system and method for object detection in 3-dimensional space uses a ground up model that makes detecting objects faster and more accurate by combining the error correction and object detection, while also using flat planes as a basis for the mathematics. Two or more 3D scanning devices, such as scanning LIDARs, collect a 3D dataset for a field of view (FOV). The FOV is divided into boxes with a small degree of overlap, then each box is processed iteratively to identify features of an object contained within the box. Box size is chosen so that the portion of the object contained within the box can be assumed to be planar. Adjoining boxes are evaluated for shared data points then merged it the form part of the same object.