Multi-Task Image Recognition Gating for Lower Vehicle Compute Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle surveillance systems require significant computational resources for accurate image recognition across multiple cameras, leading to high calculation loads without a corresponding reduction in recognition accuracy.
Innovation Solution
An information processing apparatus that decides whether to perform additional recognition tasks based on the results of initial semantic segmentation, dynamically generating parameters for neural networks to process only necessary image areas and features, thereby reducing overall calculation without degrading accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If similar recognition processing is always performed for images from multiple cameras to ensure high recognition accuracy, then recognition accuracy is improved, but calculation amount increases
Solution Approach 1:
The patent performs semantic segmentation first as a preliminary step to identify objects of interest in the image. Based on this preliminary analysis, the system decides which subsequent recognition tasks (object detection, motion detection, distance detection) should be performed. This preliminary action avoids performing all recognition processing unnecessarily, thereby reducing calculation amount while maintaining recognition accuracy for relevant objects
Solution Approach 2:
The patent dynamically adjusts the recognition processing strategy based on scene features detected from semantic segmentation. The system adaptively determines whether to perform object detection, motion detection, or distance detection by analyzing the presence and characteristics of objects in the image. This dynamic approach allows the system to optimize calculation resources according to actual scene requirements while ensuring accurate recognition when needed
2Reliability
If all recognition tasks are performed for every image to maintain comprehensive surveillance, then recognition completeness is improved, but power consumption increases
Solution Approach 1:
The patent performs semantic segmentation as a preliminary filtering step that identifies objects of interest before committing to more energy-intensive recognition tasks. This preliminary action ensures that comprehensive surveillance is maintained for relevant objects while avoiding unnecessary processing for irrelevant areas, thus balancing recognition completeness with power consumption
Solution Approach 2:
The patent applies different recognition processing strategies to different regions of the image based on the presence of objects of interest. For regions containing objects of interest, comprehensive recognition tasks are performed. For regions without objects of interest, processing is reduced or skipped. This local quality approach maintains recognition completeness where needed while reducing power consumption in irrelevant areas
Data Source
AI summary
An information processing apparatus according to the present technology includes a processing section. The processing section is capable of processing a plurality of tasks for a recognition target, including first and second tasks that share a feature extraction. The processing section decides whether or not to perform the second task processing using a recognition result of the recognition target from the first task processing.