Occluded Object Detection via Temporal Tracking and Model Update
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face difficulties in detecting objects that are occluded by foreground objects, leading to incomplete tracking and operation of occluded objects.
Innovation Solution
An information processing apparatus with a determination unit, tracking unit, labeling unit, learning unit, and detection unit that updates the detection model by identifying and labeling occluded objects in previous images, allowing for improved detection of occluded objects of the same class in subsequent images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a detection model is used to detect objects from images, then detection speed is improved, but detection accuracy of occluded objects deteriorates
Solution Approach 1:
The system performs preliminary actions by tracking objects across multiple frames and identifying occlusion states before detection failure occurs. When an object is detected as occluded in current frame, the system proactively searches for it in previous frames and prepares recovery strategies, preventing complete loss of track rather than reacting after detection fails.
Solution Approach 2:
The system establishes feedback loops where detection results from current frame inform search strategies in previous frames. When occlusion is detected, the system feeds back this information to adjust the search scope and intensity in historical frames, and uses the recovered object information to refine future detection performance, creating a closed-loop improvement system.
2Measurement precision
If the detection model is updated with labeled occluded objects, then detection accuracy of occluded objects is improved, but device complexity increases
Solution Approach 1:
The system performs self-service by automatically generating training data through its own tracking and occlusion detection capabilities. Instead of requiring external manual annotation, the system identifies occluded objects during normal operation, retrieves them from historical frames, and uses these self-generated samples to retrain and improve its own detection model, creating a self-improving cycle.
Solution Approach 2:
The system changes parameters dynamically by adjusting the scope of historical frame search, the threshold for occlusion detection, and the frequency of model updates based on operational conditions. These parameter adjustments allow the system to balance between computational complexity and detection accuracy, adapting to different scenarios without requiring complete system redesign.
Data Source
AI summary
An information processing apparatus includes: a determination unit that determines that a first object has been detected by a detection model from a first image included in a series of time-series images, the first object having not been detected from one or more second images that are included in the time-series images and chronologically earlier than the first image; a tracking unit that extracts, from the one or more second images, the first object that has been occluded by a second object under a predetermined condition; a labeling unit that adds a label to the first object extracted from the one or more second images; a learning unit that learns the one or more second images including the first object to which the label is added; an update unit that updates the detection model on the basis of a learning result of the learning unit; and a detection unit that detects, from a third image, a third object occluded under the predetermined condition by executing the updated detection model.