Adaptive Lost State Determination in Image Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current tracking techniques using Deep Neural Networks (DNNs) face challenges in maintaining reliable tracking when the posture and motion of a target change significantly, often leading to incorrect determination of the lost state due to the absence of consideration for object-specific prediction reliability and ambient conditions.
Innovation Solution
An image processing apparatus and method that acquire likelihood information from both tracking and detection techniques to generate a dynamic determination reference for identifying the lost state, incorporating motion estimation parameters and likelihood thresholds to differentiate between tracking targets and ambient objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a common predetermined reference is used for lost state determination regardless of object and ambient situation, then the determination process is simple, but the tracking reliability deteriorates due to incorrect lost state determination
Solution Approach 1:
The patent applies dynamics by making the determination reference adaptive rather than static. The reference is dynamically adjusted based on the detection result, specifically using the difference between the detection likelihood of the tracking target and other objects. This allows the system to adapt to varying ambient situations and object characteristics, resolving the contradiction between operational simplicity and tracking reliability.
Solution Approach 2:
The patent changes the parameter of the determination reference from a fixed predetermined value to a variable value derived from detection likelihood differences. By calculating the reference as the difference between the likelihood of the tracking target and other objects in the detection result, the system optimizes the threshold for lost state determination according to actual conditions, thereby improving reliability while maintaining reasonable complexity.
2Productivity
If the lost state is determined using a common reference, then the processing is efficient, but the template updating is excessively performed causing tracking failure
Solution Approach 1:
The patent optimizes the determination reference parameter by using detection likelihood differences. This dynamic parameter adjustment prevents both excessive template updating and failure to update when needed, thereby maintaining tracking stability while avoiding unnecessary processing overhead.
Solution Approach 2:
The system uses feedback from the detection result to adjust the determination reference. By continuously monitoring the difference between tracking target likelihood and other object likelihoods, the system adapts the lost state determination threshold in real-time, preventing excessive template updates and improving tracking stability.
3Reliability
If the detector is used as an auxiliary unit for lost state determination, then the tracking robustness is improved, but the device complexity increases
Solution Approach 1:
The patent applies universality by making the detection unit serve multiple functions: both object detection and provision of reference information for lost state determination. The detection likelihood values obtained during normal detection operations are reused to dynamically adjust the lost state determination reference, eliminating the need for separate auxiliary mechanisms and reducing overall system complexity while improving robustness.
Data Source
AI summary
An image processing apparatus comprising one or more memories storing instructions and one or more processors that execute the instructions to acquire first information including likelihoods of tracking targets in a plurality of first candidates of the tracking target in a target image by using a tracking technique for tracking an object in an image, acquire second information including likelihoods of tracking targets in a plurality of second candidates of the tracking target in the target image by using a detection technique for detecting an object in an image, generate a determination reference for determining, based on the first information and the second information, whether a state is a lost state in which it is not possible to specify a tracking target, and determine, using the determination reference, whether the target image is set in a lost state.