Center-Point Object Detection for Noise-Robust Bounding Boxes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques in autonomous vehicles face challenges in accurately generating bounding boxes for objects using sensor data, leading to inefficiencies and potential safety issues due to noise susceptibility and resource-intensive processing.
Innovation Solution
The use of unimodal confidence values and masking techniques to determine candidate center points and pixels, reducing the number of candidate bounding boxes through non-maximum suppression, and training models to propagate accurate object data parameters for improved detection and tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional object detection techniques are used to generate bounding boxes from sensor data, then detection coverage is comprehensive, but accuracy is reduced due to noise susceptibility and processing efficiency deteriorates due to resource-intensive operations
Solution Approach 1:
The patent extracts and focuses computation on the most informative region - the center point of objects. By applying center-based detection, the system extracts only the essential detection information from the center region rather than processing all pixels uniformly, thereby improving both accuracy and efficiency
Solution Approach 2:
The patent applies different processing strategies to different regions of the detection space. High-resolution processing is applied to the center region where precision is critical, while peripheral regions use lower-resolution processing. This local quality differentiation improves overall efficiency without sacrificing critical detection accuracy
2Reliability
If comprehensive pixel processing is performed for object detection, then detection coverage is complete, but processing load increases and noise susceptibility worsens
Solution Approach 1:
The patent extracts only the essential detection information from the center point of objects, eliminating the need to process all pixels with equal complexity. This extraction approach reduces processing complexity while maintaining detection reliability by focusing on the most informative region
Solution Approach 2:
Instead of processing from the outside in (traditional approach), the patent inverts the approach by processing from the center outward. This inversion allows the system to first establish accurate center points with simple processing, then expand outward only as needed, reducing overall processing complexity
Data Source
AI summary
Techniques for detecting and tracking objects in an environment are discussed herein. For example, techniques can include detecting a center point of a block of pixels associated with an object. Unimodal (e.g., Gaussian) confidence values may be determined for a group of pixels associated with an object. Proposed detection box center points may be determined based on the Gaussian confidence values of the pixels and an output detection box may be determined using filtering and/or suppression techniques. Further, a machine-learned model can be trained by determining parameters of a center pixel of the detection box and a focal loss based on the unimodal confidence value which can then be backpropagated to the other pixels of the detection.