Queued Radar Frame Feature Extraction for Sparse Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing radar systems struggle to accurately identify both large and small objects, such as cars and pedestrians, due to sparse radar detections in individual frames, leading to misidentification and inefficiencies in advanced driver assistance and autonomous driving applications.
Innovation Solution
A method involving time-ordered queuing of radar frames, farthest point sampling across multiple frames, and hierarchical temporal-spatial encoding to group and extract features, using a modified form of farthest point sampling that considers both spatial and temporal offsets, along with cascaded encoder stages for enhanced feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If radar systems process individual frames independently, then processing speed is maintained, but object identification accuracy deteriorates due to sparse detections
Solution Approach 1:
The system performs preliminary actions by queuing multiple radar frames in advance and performing farthest point sampling across the queue before feature extraction. This preliminary organization of temporal data allows the network to access historical detection information without requiring real-time processing of each frame individually, thereby improving identification accuracy while maintaining processing efficiency.
Solution Approach 2:
The invention transitions from processing single 2D spatial frames to processing 3D temporal-spatial queues of frames. By adding the temporal dimension through queuing multiple frames and using farthest point sampling that considers both spatial and temporal offsets, the system creates denser object representations without proportionally increasing processing complexity.
2Measurement precision
If farthest point sampling is performed across multiple frames, then object representation density improves, but computational complexity increases
Solution Approach 1:
The system applies partial action by performing farthest point sampling on a limited queue of frames (e.g., 3-5 frames) rather than all available frames. This selective sampling across a constrained temporal window achieves sufficient object representation density while limiting the computational burden to a manageable scope.
Solution Approach 2:
The farthest point sampling algorithm applies local quality by considering both spatial offsets and temporal offsets when selecting sample points. This localized approach focuses computational resources on identifying representative points within specific spatial-temporal regions rather than uniformly processing all points, thereby improving representation density efficiently.
3Measurement precision
If temporal search range is increased to capture more object points, then detection completeness improves, but processing time increases
Solution Approach 1:
The system implements dynamic temporal search range adjustment based on object speed. For slow-moving objects, a larger temporal search range is applied to capture sufficient points for accurate representation. For fast-moving objects, the temporal range is reduced to match the object's displacement, preventing excessive processing while maintaining detection completeness.
Solution Approach 2:
The invention dynamically changes the temporal search range parameter based on detected object speed and other conditions. This parameter adaptation allows the system to optimize the balance between detection completeness and processing time for different operational scenarios, rather than using a fixed temporal window.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A computerized technique is disclosed of identifying object features in an environment of a vehicle. The technique includes receiving, by an encoder, data representing a plurality of frames, the frames providing point-in-time versions of a segmented pointed cloud derived from output of one or more radar sensors of the vehicle and including points that represent radar detections corresponding to an object in the environment at respective instants in time. The technique further includes arranging the plurality of frames in a time-ordered queue and processing the frames in the queue, including (i) selecting, from among the points, a plurality of sample points that spans multiple frames of the queue, (ii) forming a plurality of groups of points based on respective sample points of the plurality of sample points, and (iii) extracting features of the object based on the plurality of sample points and the plurality of groups.