Vehicle Object Recognition Using Confidence-Triggered Extra Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition systems in vehicles, especially for autonomous driving, face challenges in reliably identifying objects with low confidence levels, leading to potential inaccuracies in object classification, particularly when the vehicle is in an uncommon relative position to the object.
Innovation Solution
A method that involves capturing images using a camera unit, performing an initial object recognition procedure, and requesting additional images from a central control unit if the confidence level is below a predefined threshold, allowing for a second recognition procedure based on the initial and additional images to improve recognition reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single image is used for object recognition, then the processing time is short, but the recognition reliability is low when the vehicle is in uncommon relative positions
Solution Approach 1:
The system performs preliminary object detection and confidence assessment on the first captured image before deciding whether to request additional images. This preliminary action allows the system to only acquire more images when necessary, improving reliability without always incurring the time cost of multiple image acquisitions.
Solution Approach 2:
The system dynamically adjusts the number of images to be processed based on the confidence level of the initial recognition. When confidence is high, only one image is used; when confidence is low, additional images are requested. This dynamic approach optimizes the balance between recognition reliability and processing time.
2Measurement precision
If additional images are requested from central control unit, then recognition accuracy improves, but system complexity increases
Solution Approach 1:
The central control unit acts as an intermediary that receives object information requests from the vehicle's data processing apparatus and provides additional images in response. This intermediary architecture allows the vehicle to access multiple images without directly managing the complexity of image acquisition and coordination.
Solution Approach 2:
The system uses feedback from the confidence level assessment to determine whether to request additional images. The confidence level serves as feedback that triggers or prevents further image acquisition, creating a closed-loop system that improves accuracy while avoiding unnecessary complexity.
3Reliability
If multiple recognition procedures are performed iteratively, then object identification reliability improves, but processing time increases
Solution Approach 1:
The system performs preliminary confidence assessment on the first image before initiating iterative recognition procedures. This preliminary check ensures that iterative processing is only started when necessary, reducing overall processing time while maintaining high reliability when multiple procedures are needed.
Solution Approach 2:
The number of iterative recognition procedures is dynamically determined based on the confidence level of each procedure. The system continues iterating only as long as the confidence level remains below the threshold, automatically stopping when sufficient reliability is achieved or the maximum number of procedures is reached.
Data Source
Figure 1~3(b)
Figure 4~5
Figure 6(a)~6(b)
AI summary
The disclosure relates to method for recognizing an object (38) in the surroundings of a vehicle (10). The method comprises capturing at least one image (I0) of the surroundings. Subsequently, a first object recognition procedure is performed and a first object information request (IR1) is provided to a central control unit (40), if the object recognition procedure returns a recognition result and an associated confidence information being inferior to a predefined confidence threshold. Moreover, the method comprises receiving at least one additional image (I1, I2, I3) from the central control unit (40) and performing a second object recognition procedure based thereon. The disclosure also relates to a method for controlling a vehicle (10), comprising performing the above method for recognizing an object (38). Furthermore, a corresponding data processing apparatus (14), computer program (24), and computer-readable storage medium (22) are described. Also a vehicle (10) having such a data processing apparatus (14) is presented. Additionally, a method for providing an object information (OI) to a vehicle (10), a corresponding data processing apparatus (42), computer program (50), and computer-readable storage medium (48) are disclosed. Moreover, an object recognition system (58) is explained.