3D Vehicle Model Tracking for Lateral Position Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing adaptive cruise control (ACC) and forward collision warning (FCW) systems using radar sensors struggle with precise lateral tracking, particularly in situations involving neighboring lane disturbances, vehicle overtaking, sharp turns, and 'late' vehicle cut-ins, due to limitations in detecting vehicles without seeing their entire backside and maintaining tracking when parts of the vehicle move out of the camera's field of view.
Innovation Solution
The implementation of a three-dimensional model-based system using two-dimensional data from cameras to track vehicles, allowing for accurate lateral position detection and turning intention recognition, even when only a side of the vehicle is visible, and enabling seamless tracking across different vehicle states and orientations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a camera is used to detect vehicle lateral position and provide lane position information, then lateral tracking capability is improved, but the camera can only track vehicles when most or the entire backside of the vehicle is within the camera's field of view, causing tracking loss in cut-in situations and overtaking scenarios
Solution Approach 1:
The patent transitions from 2-D camera images to a 3-D vehicle model representation. By constructing a three-dimensional model of the vehicle with multiple surfaces (front, back, left side, right side), the system can track vehicles in various orientations and positions, not just when the backside is visible. This dimensional transformation allows the system to maintain tracking capability across cut-in situations, overtaking scenarios, and sharp turns where the vehicle orientation changes significantly.
2Measurement precision
If radar sensors are used for longitudinal distance and velocity measurements, then longitudinal measurement accuracy is improved, but radar systems cannot detect certain traffic situations relying on precise lateral tracking, such as neighboring lane disturbances and late vehicle cut-ins
Solution Approach 1:
The patent combines radar sensor data with camera-based 3-D vehicle modeling to create a hybrid tracking system. The radar provides accurate longitudinal distance and velocity measurements, while the camera system with 3-D modeling adds lateral position detection and orientation tracking capabilities. By merging these two sensing modalities, the system achieves both precise longitudinal measurement and improved lateral tracking, enabling detection of neighboring lane disturbances, overtaking scenarios, and late cut-ins that radar alone cannot detect.
3Ease of operation
If existing monochrome cameras are used for vehicle detection, then vehicle detection is possible, but the cameras need to see most or the entire backside of a vehicle to detect and track it, causing tracking failure when vehicles turn sharply or cut in late
Solution Approach 1:
The system performs preliminary construction of a complete 3-D vehicle model that includes all surfaces (front, back, left side, right side) before tracking begins. This pre-established model allows the system to predict and track vehicle positions even when certain surfaces move out of the camera's field of view. The model is updated incrementally as different surfaces become visible, ensuring continuous tracking reliability in dynamic situations such as sharp turns and late cut-ins, where the vehicle orientation changes rapidly.
Data Source
AI summary
Systems and methods for detecting a vehicle. One system includes a controller. The controller is configured to receive images from a camera mounted on a first vehicle, identify a surface of a second vehicle located around the first vehicle based on the images, and generate a three-dimensional model associated with the second vehicle. The model includes a first plane and a second plane approximately perpendicular to the first plane. The first plane of the model is associated with the identified surface of the second vehicle. The controller is further configured to track a position of the second vehicle using the three-dimensional model after the identified surface falls at least partially outside of a field-of-view of the at least one camera.