Biased On-Device Image Classifier for Energy and Bandwidth Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods on portable devices are resource-intensive, leading to shortened battery life and high bandwidth consumption, as they often require transmitting images to cloud-based servers for processing, which increases latency and communication costs.
Innovation Solution
A two-stage image processing system where a local portable device performs low-complexity operations, such as interest point detection and feature extraction, and filters out less-relevant data, while a remote device handles high-complexity tasks like image understanding, reducing the need for extensive data transmission and conserving local resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image processing is performed on a portable device using known methods, then image understanding and analysis can be achieved, but energy consumption increases significantly and battery life is shortened
Solution Approach 1:
The image processing task is segmented into two stages: a first stage performed on the portable device using a biased classifier for quick filtering, and a second stage performed on a remote device using a full classifier for accurate understanding. This segmentation allows the portable device to conserve energy while still achieving comprehensive image analysis through the remote device.
Solution Approach 2:
A biased on-device image classifier acts as an intermediary that pre-processes images locally and selects candidate images for transmission to the remote device. This intermediary component reduces the overall energy consumption by filtering out non-candidate images before they reach the energy-intensive remote processing stage.
2Measurement precision
If all captured images are transmitted to a cloud-based server for processing, then comprehensive image analysis can be achieved, but bandwidth consumption and communication costs increase significantly
Solution Approach 1:
The system extracts only the essential feature vectors from images using interest point detection and feature extraction algorithms on the portable device. Only these extracted features and candidate image identifiers are transmitted to the remote device, rather than transmitting the complete image data, thereby significantly reducing bandwidth consumption.
Solution Approach 2:
Feature extraction and image classification are performed in advance on the portable device before transmission. This preliminary processing identifies candidate images that require further analysis, allowing the system to transmit only relevant data to the remote device and avoid unnecessary bandwidth consumption.
3Use of energy by moving object
If a biased classifier is used on the portable device to filter images, then energy consumption and bandwidth usage are reduced, but some candidate images may be misclassified as false positives
Solution Approach 1:
The system dynamically adjusts the classification threshold of the biased classifier to optimize the balance between energy efficiency and classification accuracy. By making the threshold adjustable, the system can adapt to different operational requirements, reducing false positives while maintaining energy savings.
Solution Approach 2:
The system incorporates feedback mechanisms where classification results from the biased classifier are evaluated and used to refine future classification decisions. This feedback loop helps reduce false positives over time while maintaining the energy efficiency benefits of the biased approach.
Data Source
AI summary
Examples of the disclosure enable efficient processing of images. One or more features are extracted from a plurality of images. Based on the extracted features, the plurality of images are classified into a first set including a plurality of first images and a second set including a plurality of second images. One or more images of the plurality of first images are false positives. The plurality of first images and none of the plurality of second images are transmitted to a remote device. The remote device is configured to process one or more images including recognizing the extracted features, understanding the images, and/or generating one or more actionable items. Aspects of the disclosure facilitate conserving memory at a local device, reducing processor load or an amount of energy consumed at the local device, and/or reducing network bandwidth usage between the local device and the remote device.