Object Detection Model for Partial Occlusion Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods face decreased accuracy when objects are partially blocked, as they rely on training data that may not account for partial occlusions, leading to inconsistent detection results across multiple detectors.
Innovation Solution
A computer-implemented detection method that uses a learned model generated from training data including both partial and entire images of objects, allowing for the identification of entire objects within images and determining the presence of other objects by comparing predicted images with input images, thereby increasing detection accuracy for partially blocked objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If training data for each object is used separately, then detection process is simplified, but detection accuracy decreases when objects are partially blocked
Solution Approach 1:
The patent combines multiple object detection processes into a unified detection framework. Instead of treating each object separately, the system jointly detects multiple objects in an image, allowing the detection model to understand spatial relationships and occlusions between objects. This merging approach maintains process simplicity while improving accuracy in occluded scenarios.
Solution Approach 2:
The detection system is designed to handle multiple object types and occlusion scenarios simultaneously through a universal detection framework. The same detection model can process various object combinations and occlusion patterns without requiring separate specialized detectors for each case, achieving both simplicity and accuracy.
2Measurement precision
If training data including combinations of multiple objects is used, then detection accuracy for occluded objects improves, but the number of training data required becomes enormous
Solution Approach 1:
The system employs a dynamic detection approach where the detection model adaptively adjusts to different object configurations and occlusion patterns during inference. Rather than requiring exhaustive static training data for all possible combinations, the model learns generalizable features that enable it to handle diverse scenarios with fewer training examples.
Solution Approach 2:
The detection framework utilizes parameter changes in the detection model to adapt to different object arrangements and occlusion levels. By adjusting detection parameters and using learned representations, the system can generalize to unseen object combinations without requiring enormous amounts of specific training data for each scenario.
3Reliability
If multiple detectors are prepared to detect partial objects, then occluded objects can be detected, but detection results may be inconsistent and require complex validation
Solution Approach 1:
The patent merges multiple detection functions into a single unified detection model that outputs detection results for multiple objects simultaneously. This eliminates the need for separate detectors and complex validation logic to reconcile inconsistent results, while maintaining reliable detection of occluded objects through joint processing.
Data Source
AI summary
A computer-implemented detection method includes, in response to inputting a first image including a region of one or more objects to a learned model, identifying a first entire image corresponding to entirety of a first object as a detection candidate, the learned model being generated by learning training data including an image corresponding to a part of an object and an entire image corresponding to entirety of the object, detecting an existing region of the first target object in the first image in accordance with a comparison between the identified first entire image and the region of the one or more target objects, and determining, based on a specific image obtained by invalidating the existing region in the first image, whether another target object is included in the first image.