2D Road Object Detection With Centroid Offset Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Object detection using LiDAR sensors is costly and prone to misclassification and false positives, leading to unreliable outcomes, especially at longer distances, due to the inherent reliance on expensive LiDAR scanners and the lack of effective utilization of 2D detection data.
Innovation Solution
A system and method for object detection using 2D sighting data without explicit depth sensors, employing a 2D-to-2D association mechanism to correct centroid offsets, a filtering mechanism to remove false positives, and a scaling mechanism to enhance accuracy, enabling near-real-time 3D feature detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LiDAR sensors are used for object detection, then detection accuracy is improved, but system cost increases significantly
Solution Approach 1:
The patent creates a virtual 3D point cloud by projecting 2D bounding boxes and centroids from image data into 3D space using camera intrinsics and extrinsics. This virtual 3D representation copies the essential spatial information needed for detection without requiring physical LiDAR sensors, thereby maintaining detection accuracy while eliminating the need for expensive depth sensors.
Solution Approach 2:
The patent introduces an intermediary processing stage that transforms 2D detection results into 3D spatial representations through mathematical projection. This intermediary step bridges the gap between 2D image data and 3D object detection requirements, allowing the system to achieve 3D detection capabilities through 2D sensors alone.
2Productivity
If LiDAR based classification is used, then object detection is performed, but misclassification increases at longer distances
Solution Approach 1:
The patent implements a feedback mechanism where 2D detection results are used to generate initial 3D hypotheses, which are then refined through iterative projection and validation. The system continuously refines object hypotheses by projecting 2D bounding boxes into 3D space and validating against spatial constraints, improving classification reliability at longer distances through this closed-loop feedback process.
Solution Approach 2:
The patent transitions from 2D image space to 3D spatial reasoning by projecting 2D detection results into 3D point clouds. This dimensionality change allows the system to leverage depth information derived from geometric projection, improving classification reliability at longer distances by incorporating spatial context that is unavailable in pure 2D image analysis.
3Device complexity
If 2D sensors are used without depth sensing, then system cost is reduced, but false positives increase
Solution Approach 1:
The patent performs preliminary 2D object detection and bounding box estimation before projecting into 3D space. By pre-processing the 2D detection results and generating initial spatial hypotheses, the system establishes a foundation for accurate 3D localization that reduces false positives. This preliminary action ensures that only plausible object hypotheses are carried forward into the 3D projection stage.
Solution Approach 2:
The patent employs dynamic hypothesis generation where multiple 3D object hypotheses are created and evaluated based on their consistency with 2D projections and spatial constraints. The system dynamically adjusts and refines object hypotheses through iterative validation, eliminating inconsistent hypotheses and retaining only those that satisfy geometric and spatial constraints, thereby reducing false positives.
Data Source
AI summary
The disclosure provides a method, a system, and a computer program product for object detection using 2D sighting data of the object obtained using a 2D sensor. The method comprises obtaining 2D sighting data of the object using the 2D sensor. Further, the method comprises determining position candidate data for the object based on (i) 2D centroid data associated with the 2D sighting data of the object and (ii) 3D centroid data determined using projection data of one or more skew lines associated with the 2D centroid data. The position candidate data is then filtered based on (i) offset data associated with vector offset between the 2D centroid data and the 3D centroid data, (ii) postprocessing data, and (iii) scaling factor data associated with the position candidate data. Further, the detection data for the object is outputted based on the filtered position candidate data.