LiDAR Feature Vector Update for Real-Time Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current real-time object detection systems in unmanned vehicles face delays due to the time-consuming process of collecting and processing LiDAR point cloud data, leading to low real-time performance and efficiency.
Innovation Solution
The method involves acquiring target point cloud data within a preset angle less than the full angle, determining one-dimensional LiDAR point features, and updating a previous frame of LiDAR feature vector to generate a current frame vector for real-time object detection, thereby reducing data delay and improving processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If point cloud data is collected by a LiDAR device rotating around to perform object detection, then the detection accuracy is improved, but the data collection time increases leading to low real-time performance
Solution Approach 1:
The patent divides the full 360-degree point cloud data into multiple smaller angular segments (e.g., 0-90 degrees, 90-180 degrees, etc.). Each segment is processed independently and simultaneously, allowing parallel computation. This segmentation reduces the time required to collect and process each segment while maintaining comprehensive coverage through concatenation of results from all segments.
2Reliability
If the LiDAR device rotates around to collect complete point cloud data, then the recall rate is improved, but the overall running time increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing and extracting features from each angular segment independently before the complete dataset is available. By processing segments in parallel and concatenating results, the system achieves faster overall processing while maintaining high recall rates through comprehensive coverage of all angular regions.
3Reliability
If point cloud data from the entire angle range is processed, then the detection completeness is improved, but the algorithm running time increases
Solution Approach 1:
The patent segments the point cloud data by angular range and processes each segment independently through feature extraction and object detection. The results from all segments are then concatenated to form the complete detection result. This approach enables parallel processing of segments, significantly reducing algorithm running time while maintaining detection completeness through comprehensive coverage of all angular regions.
Data Source
AI summary
A method, an apparatus and an electronic device for real-time object detection are provided according to the present disclosure. In the method, target point cloud data in a range of a preset angle is acquired, where the target point cloud data includes one or more data points, and the preset angle is less than the round angle. A one-dimensional LiDAR point feature of each of the data points in the target point cloud data is determined. A previous frame of LiDAR feature vector is updated based on the one-dimensional LiDAR point feature of each of the data points in the target point cloud data to generate a current frame of LiDAR feature vector. Object information is determined in a real-time manner based on the current frame of LiDAR feature vector. With the method, apparatus and electronic device for real-time object detection provided according to embodiments of the present disclosure, a part of the point cloud data is selected to be processed in real time, which can reduce data delay, reduce the overall running time, and improve processing efficiency. By updating the whole LiDAR feature vector, the object information can be determined in real time in the whole range, so as to realize the object detection in the whole range.