Convolution Model for Mobile Object Detection and Segmentation

Solution Overview

Problem

Existing mapping applications on mobile computing devices face challenges in low power and processing environments, particularly when processing camera images in real-time, leading to high bandwidth and battery consumption.

Innovation Solution

A convolution model is implemented on mobile devices to detect objects and segment environmental blocks, featuring encoding and decoding layers with skip connections, and linknet blocks, allowing for efficient object detection and segmentation while minimizing computational power and battery usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If mapping applications process camera images in real time on mobile computing devices, then object detection and segmentation functionality is improved, but battery power consumption increases

Engineering Contradiction:

Improvereal-time image processing capabilityVSAvoidbattery power consumption

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The neural network is divided into encoding layers and decoding layers with specific blocks (detector block heads, linknet blocks, up-sampling layers) that process different aspects of image analysis separately. This segmentation allows the system to perform multiple detection tasks efficiently while managing computational resources on mobile devices.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The convolution model performs both object detection and environmental block segmentation using a unified architecture with shared components (encoding layers, skip connections). This multi-functional approach reduces redundant computation compared to running separate models for each task, thereby lowering overall power consumption.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Productivity

If mapping applications process camera images in real time on mobile computing devices, then object detection and segmentation functionality is improved, but processing power requirements increase

Engineering Contradiction:

Improvereal-time image processing capabilityVSAvoidprocessing power

Core Design Contradiction:

ProductivityVSPower

Solution Approach 1:

The neural network is divided into encoding layers and decoding layers with specific blocks (detector block heads, linknet blocks, up-sampling layers) that process different aspects of image analysis separately. This segmentation allows the system to perform multiple detection tasks efficiently while managing computational resources on mobile devices.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The convolution model combines object detection and environmental block segmentation into a single unified processing pipeline. By merging these tasks and sharing computational components (encoding layers, skip connections), the system reduces overall processing requirements compared to running separate models.

Inventive Principle:

Principle #5Merging (Combining)

3Measurement precision

If mapping applications use convolution models with multiple layers and skip connections, then detection accuracy is improved, but device complexity increases

Engineering Contradiction:

Improveobject detection accuracyVSAvoidmodel architecture complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The neural network is divided into encoding layers and decoding layers with specific blocks (detector block heads, linknet blocks, up-sampling layers) that process different aspects of image analysis separately. This segmentation allows the system to perform multiple detection tasks efficiently while managing computational resources on mobile devices.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The model uses a subset of encoding layers and decoding layers with selective skip connections rather than a complete deep architecture. This partial action approach provides sufficient detection accuracy for mobile applications while reducing the overall complexity and computational burden compared to full-scale models.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS11593593B2Low power consumption deep neural network for simultaneous object detection and semantic segmentation in images on a mobile computing device

Publication Date: 2023.02.28 MAPBOX INC

AI summary

A mobile computing device receives an image from a camera physically located within a vehicle. The mobile computing device inputs the image into a convolutional model that generates a set of object detections and a set of segmented environment blocks in the image. The convolutional model includes subsets of encoding and decoding layers, as well as parameters associated with the layers. The convolutional model relates the image and parameters to the sets of object detections and segmented environment blocks. A server that stores object detections and segmented environment blocks is updated with the sets of object detections and segmented environment blocks detected in the image.