Parallel Neural Network Computation for Embedded Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computation techniques for object tracking using neural networks face challenges in maintaining accuracy due to feature differences in objects across varying environments, leading to increased processing time and reduced frame rates when online learning is applied, especially in embedded devices with limited computation performance.
Innovation Solution
A computation apparatus and method that enables parallel execution of neural network computation and online learning by separating tasks into offline and online layers, allowing for simultaneous processing of neural network features and coefficient updates, thereby reducing memory access conflicts and processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If online learning is applied to improve object tracking accuracy, then tracking accuracy is improved, but processing time per frame increases and frame rate deteriorates
Solution Approach 1:
The patent segments the neural network into two distinct parts: an offline layer with fixed coefficients and an online layer with updateable coefficients. This segmentation allows the offline layer to be executed once and reused across multiple frames, while only the online layer requires repeated execution for each frame's tracking updates. By dividing the network this way, the system achieves both accurate tracking through online learning and maintained frame rates through efficient resource utilization.
Solution Approach 2:
The patent performs preliminary computation by calculating the offline layer features once and storing them for reuse across multiple frames. This preliminary action eliminates the need to recompute these features for every frame, significantly reducing the processing time required per frame while maintaining tracking accuracy through subsequent online learning updates.
2Measurement precision
If online learning processing is incorporated into the computation apparatus, then tracking accuracy is improved, but the processing time period per frame increases
Solution Approach 1:
The patent enables continuous useful action by allowing the offline layer computation to be performed once and then reused continuously across multiple frames without requiring re-computation. Simultaneously, the online layer performs continuous updates using this cached offline information. This continuity eliminates redundant computations and maintains both high inference accuracy and reduced processing time per frame.
3Measurement precision
If neural network computation and online learning are executed sequentially, then computation accuracy is maintained, but processing efficiency decreases
Solution Approach 1:
The patent segments the computation into independent offline and online layers that can be executed in parallel rather than sequentially. The offline layer processes features that remain constant across frames, while the online layer handles frame-specific updates. This segmentation allows both computations to proceed simultaneously, maintaining computational accuracy while dramatically improving processing efficiency.
Solution Approach 2:
The patent merges the offline and online layer computations into a unified neural network architecture where both layers work together on the same input data. This merging allows the system to maintain the accuracy benefits of complete neural network processing while achieving parallel execution efficiency, as both layers process information concurrently rather than requiring separate sequential passes.
Data Source
AI summary
A computation apparatus, comprises a first processing unit configured to obtain a first feature by executing computation of a neural network with use of a first coefficient that is not to be updated in online learning of the neural network, a second processing unit configured to obtain a second feature by executing the computation of the neural network with use of the first feature and a second coefficient that is to be updated in the online learning, and an update unit configured to update the second coefficient by executing the online learning with use of the second coefficient and a second feature that has been obtained by the second processing unit in a past. Processing of the first processing unit and processing of the update unit are executed in parallel.