Vehicle Occlusion Detection Through Frame Tracking and Dual Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection frameworks for autonomous vehicles fail to accurately and efficiently detect occluded objects, which is crucial for safe operation and motion planning, as they cannot recover the shape of occluded objects or effectively handle occlusion detection problems.
Innovation Solution
A system and method for vehicle occlusion detection that involves training two machine learning classifiers: one for static images and another for image sequences, using extracted features to determine the occlusion status of vehicles by tracking objects across frames and applying different classifiers based on tracking success.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current object detection frameworks are used, then detection speed is maintained, but occlusion detection accuracy deteriorates
Solution Approach 1:
The patent segments the detection task by creating two specialized classifiers: one for static images and another for image sequences. This segmentation allows each classifier to be optimized for its specific task, improving overall occlusion detection accuracy while maintaining efficient processing through specialized rather than general-purpose detection
Solution Approach 2:
The system dynamically selects which classifier to apply based on whether objects can be tracked across frames. By making dynamic decisions about classification approach based on tracking success, the system optimizes both accuracy for occluded objects and efficiency for visible objects
2Measurement precision
If semantic segmentation is used, then pixel-level labeling is achieved, but object-level instance discovery fails
Solution Approach 1:
The patent merges semantic segmentation with object tracking and classification. By combining pixel-level labeling from segmentation with temporal tracking and classifier-based instance identification, the system recovers both precise pixel-level information and complete object-level instance data that neither method could achieve alone
3Adaptability or versatility
If current object detection frameworks are used, then general object detection is performed, but occluded object detection fails
Solution Approach 1:
The system performs preliminary object tracking across frames before classification. By pre-processing the detection task with tracking to identify potentially occluded objects, the subsequent classification can focus specifically on determining occlusion status, improving reliability for occluded objects while maintaining framework versatility
Solution Approach 2:
Object tracking serves as an intermediary step between raw image input and final classification. This intermediary process provides temporal context and motion information that helps the classifiers distinguish occluded objects from visible ones, improving detection reliability without sacrificing framework adaptability
Data Source
AI summary
A system and method for vehicle occlusion detection is disclosed. A particular embodiment includes: receiving training image data from a training image data collection system; obtaining ground truth data corresponding to the training image data; performing a training phase to train a plurality of classifiers, a first classifier being trained for processing static images of the training image data, a second classifier being trained for processing image sequences of the training image data; receiving image data from an image data collection system associated with an autonomous vehicle; and performing an operational phase including performing feature extraction on the image data, determining a presence of an extracted feature instance in multiple image frames of the image data by tracing the extracted feature instance back to a previous plurality of N frames relative to a current frame, applying the first trained classifier to the extracted feature instance if the extracted feature instance cannot be determined to be present in multiple image frames of the image data, and applying the second trained classifier to the extracted feature instance if the extracted feature instance can be determined to be present in multiple image frames of the image data.