Parallel Header Perception Network for Real-Time Multi-Task Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current visual perception networks require multiple networks to perform various detection tasks, leading to increased computational load, power consumption, and reduced real-time detection capabilities.

Innovation Solution

A perception network architecture featuring a backbone and multiple parallel headers, where each parallel header is connected to the backbone and independently performs detection tasks using RPN, ROI-ALIGN, and RCNN modules, sharing a common backbone for feature extraction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple separate networks are used to implement different detection tasks, then each task can be performed with dedicated optimization, but the calculation amount and power consumption increase dramatically

Engineering Contradiction:
Improvedetection precisionVSAvoidpower consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent merges multiple separate detection networks into a single unified network architecture that shares a common backbone for feature extraction. Multiple task-specific detection heads are integrated into this shared backbone, allowing the system to perform multiple detection tasks simultaneously while reducing redundant computations and lowering power consumption compared to running separate networks for each task.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The unified network architecture is designed with multi-functionality, where a single backbone serves multiple detection tasks through shared feature extraction. The network can simultaneously perform different types of object detection, semantic segmentation, and other vision tasks by routing features to appropriate task-specific heads, making the system universal rather than task-specific.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Measurement precision

If multiple separate networks are used to implement different detection tasks, then each task can be performed with dedicated optimization, but the calculation amount increases

Engineering Contradiction:
Improvedetection precisionVSAvoidcalculation amount
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent combines multiple detection tasks into a single network framework where the backbone features are shared across all tasks. This merging eliminates redundant feature extraction computations that would occur if separate networks were used, significantly reducing the overall calculation amount while maintaining detection precision through task-specific detection heads.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The network is segmented into a shared backbone for common feature extraction and task-specific detection heads for individual tasks. This segmentation allows efficient resource utilization where the expensive feature extraction portion is shared, while only the lighter detection head computations are task-specific, optimizing the balance between precision and computational efficiency.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If multiple separate networks are used to implement different detection tasks, then each task can be performed with dedicated optimization, but the detection speed decreases making real-time detection difficult

Engineering Contradiction:
Improvedetection precisionVSAvoiddetection speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The patent merges multiple detection tasks into a single unified network that processes input images once through a shared backbone, then branches to multiple task-specific detection heads. This approach performs all detection tasks in parallel from the same extracted features, achieving real-time detection speeds that would be impossible with sequential processing through multiple separate networks.

Inventive Principle:
Principle #5Merging (Combining)

4Measurement precision

If multiple separate networks are used to implement different detection tasks, then each task can be performed with dedicated optimization, but the system complexity increases

Engineering Contradiction:
Improvedetection precisionVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent combines multiple detection tasks into a single unified network architecture, reducing system complexity by eliminating the need to manage, deploy, and coordinate multiple separate networks. The unified structure simplifies the system while maintaining detection precision through task-specific detection heads that share a common backbone.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4542444A1Object recognition method and apparatus
Publication Date: 2025.04.23 HUAWEI TECH CO LTD
  • EP4542444A1 patent drawingFigure 1
  • EP4542444A1 patent drawingFigure 2
  • EP4542444A1 patent drawingFigure 3

AI summary

This application relates to the field of artificial intelligence, and specifically, to the field of computer vision, and discloses a perception network based on a plurality of headers. The perception network includes a backbone and the plurality of parallel headers. The plurality of parallel headers are connected to the backbone. The backbone is configured to receive an input image, perform convolution processing on the input image, and output feature maps, corresponding to the image, that have different resolutions. Each of the plurality of parallel headers is configured to detect a task object in a task based on the feature maps output by the backbone, and output a 2D box of a region in which the task object is located and confidence corresponding to each 2D box. Each parallel header detects a different task object. The task object is an object that needs to be detected in the task. Higher confidence indicates a higher probability that the task object corresponding to the task exists in the 2D box corresponding to the confidence.