Adaptive Object Tracking via Dynamic Crop Region Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking systems face challenges in accurately tracking objects with varying movement speeds, as they often require adjusting the crop region size to minimize information loss or amplification, which can lead to decreased tracking accuracy or failure.
Innovation Solution
A processor-implemented method that estimates dynamic object characteristics in input images, adjusts the crop region size based on movement thresholds, and selects appropriate neural network models for object tracking, using more weight kernels and smaller pooling windows for finer analysis when the crop region is smaller.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the crop region size is increased to capture more object information during fast movement, then tracking reliability is improved, but information loss occurs due to excessive background inclusion and feature dilution
Solution Approach 1:
The patent implements dynamic adjustment of crop region size based on object movement characteristics. The system estimates object dynamics from frame sequences and adaptively modifies the crop region dimensions - expanding during fast movement to maintain object visibility and contracting during slow movement to preserve feature quality. This dynamic adaptation resolves the contradiction by making the crop region size responsive to actual tracking needs rather than fixed.
2Manufacturing precision
If the crop region size is decreased to maintain feature quality during slow movement, then manufacturing precision of feature extraction is improved, but tracking reliability deteriorates due to insufficient context
Solution Approach 1:
The system dynamically adjusts crop region size based on estimated object movement characteristics. During slow movement phases, the crop region is reduced to maintain high feature extraction precision, while during fast movement, it is expanded to ensure sufficient object context and tracking reliability. This dynamic behavior resolves the contradiction between feature precision and tracking reliability.
3Device complexity
If a single neural network model is used for all crop region sizes, then device complexity is reduced, but measurement precision of object tracking deteriorates
Solution Approach 1:
The patent employs multiple neural network models with different architectural characteristics optimized for specific crop region sizes. A first model is designed for smaller crop regions with characteristics suited to compact feature processing, while a second model is designed for larger crop regions. The system selects the appropriate model based on the current crop region size, ensuring optimal tracking precision for each scenario while managing complexity through selective model deployment.
4Measurement precision
If multiple neural network models are selected based on crop region size, then measurement precision of tracking is improved, but device complexity increases
Solution Approach 1:
The system uses multiple specialized neural network models optimized for different crop region sizes, with each model having architectural characteristics suited to its intended input size. This specialization improves tracking precision by matching model characteristics to actual operating conditions. The complexity increase is managed through conditional model selection based on crop region size rather than maintaining all models simultaneously active.
Solution Approach 2:
The patent changes the parameter of neural network model selection based on crop region size. Different models are selected according to the size category of the current crop region, allowing the system to adapt its processing characteristics to match the input data scale. This parameter-based model selection improves precision while managing complexity through structured decision-making.
Data Source
AI summary
Disclosed is a method and apparatus for adaptive tracking of a target object. The method includes method of tracking an object, the method including estimating a dynamic characteristic of an object in an input image based on frames of the input image, determining a size of a crop region for a current frame of the input image based on the dynamic characteristic of the object, generating a cropped image by cropping the current frame based on the size of the crop region, and generating a result of tracking the object for the current frame using the cropped image.