Multihead Deep Learning for 3D Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems require multiple sensors and complex processing to generate a 3D model of a vehicle's environment, which increases processing power and time, and may not accurately identify objects without multiple inputs.
Innovation Solution
A method using a monocular camera to capture 2D images and process them through multi-head deep learning to differentiate between objects and traversable space, reducing the need for multiple sensors and enhancing object detection accuracy by assigning values for heading, depth, and motion, thereby generating a 3D model for driver assistance features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple sensors and complex processing systems are used to generate 3D models, then measurement precision and reliability improve, but device complexity and processing power requirements increase
Solution Approach 1:
The patent combines multiple detection functions (object detection, traversable space identification, depth estimation, and 3D model generation) into a single integrated deep learning model that processes monocular camera images, eliminating the need for separate sensors and processing systems for each function
Solution Approach 2:
The deep learning model performs multiple functions simultaneously including detecting objects, identifying traversable space, estimating depth, and generating 3D models from a single monocular camera input, making the system multi-functional without requiring additional sensors
2Reliability
If multiple sensors and complex processing are used, then object detection reliability improves, but processing time and computational power increase
Solution Approach 1:
The patent merges sequential processing steps (object detection, space identification, depth estimation) into a single parallel deep learning inference process that operates on monocular images, significantly reducing total processing time while maintaining reliability
Solution Approach 2:
The system performs preliminary learning and feature extraction during the training phase, enabling the model to make accurate predictions with minimal computational effort during real-time operation, thus reducing processing time without sacrificing reliability
3Measurement precision
If multiple sensors are deployed to capture environmental data, then measurement precision improves, but use of energy and device complexity increase
Solution Approach 1:
The patent extracts depth and 3D spatial information from 2D monocular images through deep learning processing, eliminating the need for energy-consuming additional sensors like stereocameras or LIDAR while maintaining measurement precision
Solution Approach 2:
The system replaces physical sensor arrays (mechanical/optical systems) with a computational approach using deep learning models that process images algorithmically, reducing energy consumption associated with multiple physical sensors
Data Source
AI summary
Systems and methods are presented herein for generating a three-dimensional model based on data from one or more two-dimensional images to identify a traversable space for a vehicle and objects surrounding the vehicle. A bounding area is generated around an object identified in a two-dimensional image captured by one or more sensors of a vehicle. Semantic segmentation of the two-dimensional image is performed based on the bounding area to differentiate between the object and a traversable space. The three-dimensional model of an environment comprised of the object and the traversable space is generated based on the semantic segmentation. The three-dimensional model is used for one or more of processing or transmitting instructions useable by one or more driver assistance features of the vehicle.