Deep Learning Ball Detection for Occluded Object Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection algorithms, such as YOLO and Faster R-CNN, struggle to accurately detect occluded objects, leading to reduced detection accuracy when parts of objects are obscured.
Innovation Solution
A machine learning model is trained to predict the geometric center and radius of a blob of pixels in a smaller region extracted from an image, using a confidence score to classify and localize the object, such as a ball, by fitting predicted parameters to ground truth parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional object detection algorithms (YOLO, Faster R-CNN) are used, then general object detection capability is achieved, but detection accuracy of occluded objects deteriorates
Solution Approach 1:
The patent changes the detection parameters from standard bounding box coordinates to polar coordinates (distance and angle from geometric center). This parameter transformation enables the model to effectively represent occluded objects by describing their spatial relationship to the center point, improving detection accuracy for partially obscured objects while maintaining general detection capability
Solution Approach 2:
The patent introduces a new dimensional approach by using polar coordinate representation instead of traditional Cartesian bounding boxes. This dimensional change allows the model to capture occluded object characteristics through radial and angular measurements from the geometric center, resolving the contradiction between general detection and occluded object detection
2Measurement precision
If Non-Maximum Suppression (NMS) with IoU calculations is used for post-processing, then object localization is improved, but processing speed deteriorates
Solution Approach 1:
The patent extracts and removes the computationally intensive NMS and IoU calculation steps from the detection pipeline. By directly predicting distance and angle parameters during the main detection process, the model eliminates the need for separate post-processing localization steps, thereby maintaining localization accuracy while significantly improving processing speed
Data Source
AI summary
Embodiments are disclosed for object detection using deep learning. In some embodiments, a method comprises: extracting, with a machine learning model, a first region from an image; pooling, with the machine learning model, the first region to a second region that is smaller than the first region; predicting, with the machine learning model, a geometric center and radius of a blob of pixels in the second region and a confidence score associated with the predicting; and classifying, with the machine learning model, the blob of pixels as a ball based on the confidence score.