Vehicular Electronic Device Common Feature Map Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicular electronic devices face challenges in efficiently processing images from multiple cameras to support autonomous driving functions, requiring rapid computing with low computing load to derive accurate results for object detection and 3D information.
Innovation Solution
The method involves using a convolutional neural network (CNN) to generate a common feature map, which is then processed by object detection, bottom, and 3D networks to extract features, predict bounding boxes and 3D dimensions, and estimate orientations, allowing for accurate distance calculations and data selection across multiple networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple separate networks are used for object detection, bottom detection, and 3D estimation, then each function can be performed independently, but the computing load increases and processing speed decreases
Solution Approach 1:
The patent merges multiple separate detection networks into a unified network architecture that performs object detection, bottom detection, and 3D estimation simultaneously. The network shares common feature extraction layers and uses a single integrated model to generate all required outputs, thereby reducing computing load while maintaining accurate results for autonomous driving functions.
2Productivity
If a single unified network is used for all detection tasks, then computing load is reduced, but the complexity of the network architecture increases
Solution Approach 1:
The unified network is segmented into distinct functional modules: a shared feature extraction section that processes input images, an object detection module that identifies objects and generates bounding boxes, a bottom detection module that detects ground surfaces, and a 3D estimation module that calculates depth and spatial information. This modular segmentation manages architectural complexity while maintaining computing efficiency.
3Ease of manufacture
If traditional image processing methods are used, then the system is simpler to implement, but the accuracy and precision of object detection and distance estimation deteriorate
Solution Approach 1:
The patent replaces traditional mechanical image processing methods with a deep learning-based neural network system. The network automatically learns optimal feature representations and detection parameters from training data, achieving high measurement precision for object detection, bottom detection, and 3D estimation without requiring manual tuning of processing parameters.
Data Source
AI summary
Disclosed is an operation method of a vehicular electronic device, including receiving at least one image data from at least one camera installed in a vehicle, by at least one processor, generating a common feature map based on the image data using a convolutional neural network (CNN), by the at least one processor, and providing the common feature map to each of an object detection network, a bottom network, and a three dimensional network, by the at least one processor.