Parallel Image Processing for Driving Assistance Reactivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing driving assistance systems for motor vehicles, particularly those using artificial intelligence and neural networks, face limitations in reactivity and fail to sufficiently anticipate potential dangers, leading to situations like late braking, and current solutions such as combining with other sensors or modifying algorithms are costly and not always feasible.
Innovation Solution
A method and system that processes an additional image, resulting from geometric transformations like zooming, rotation, or luminosity modification of the captured image, in parallel with the original image processing, to generate additional control setpoints for longitudinal and/or lateral vehicle control, which are then merged to improve reactivity without altering the internal processing of the neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the neural network processes only the original captured image, then the processing complexity is low, but the reactivity and anticipation of dangers are insufficient
Solution Approach 1:
The image processing is segmented into multiple parallel processing streams: the original image processing and additional processed image processing. Each stream independently processes different versions of the image through the neural network, allowing the system to analyze the scene from multiple perspectives simultaneously, thereby improving reactivity without overwhelming a single processing path
Solution Approach 2:
The patent introduces an additional dimension to the processing by creating transformed versions of the original image (e.g., zoomed, rotated, cropped variants). This dimensional expansion allows the neural network to evaluate the same scene from multiple spatial perspectives, enhancing danger anticipation capability while maintaining manageable processing complexity through parallel execution
2Speed
If additional sensors are combined to improve reactivity, then the reactivity improves, but the cost increases
Solution Approach 1:
Instead of adding physical sensors, the patent creates virtual copies of the existing camera image through geometric transformations (zooming, rotating, cropping). These copied and transformed image versions are then processed in parallel by the neural network, achieving enhanced reactivity equivalent to having multiple sensors while using only the original camera hardware
Solution Approach 2:
The patent changes the parameters of the existing image (geometric transformations such as scale, rotation angle, crop region) to generate multiple processing variants from a single image source. This parameter manipulation allows the system to extract more information from the same hardware input, improving reactivity without the cost of additional sensors
3Speed
If the neural network algorithm is modified to improve reactivity, then the reactivity improves, but the implementation complexity and cost increase
Solution Approach 1:
The patent performs preliminary actions by pre-processing the original image into multiple transformed versions (zoomed, rotated, cropped) before they enter the neural network. This preparation work is done in parallel using standard image processing techniques, allowing the existing neural network algorithm to operate on enriched input data without requiring complex modifications to its internal structure
Solution Approach 2:
The transformed images act as intermediaries between the original camera input and the neural network processing. These intermediate processed images carry enhanced spatial information that bridges the gap between simple single-image input and the need for multi-perspective analysis, allowing the neural network to improve reactivity without direct algorithmic complexity increases
Data Source
Figure 1~3
Figure 4~6
AI summary
The invention relates to a driving assistance system (3) for longitudinal and/or lateral control of a motor vehicle, comprising an image processing device (31a) previously trained according to a learning algorithm and configured to generate as output an instruction (Scom1) for controlling the motor vehicle from an image (Im1 provided as input and captured by an on-board digital camera (2); a digital image processing module (32) configured to provide at least one additional image (Im2) as the input of an additional device (31b) identical to the device (31a), for parallel processing of the image (Im1 captured by the camera (2) and said at least one additional image (Im2), so that said additional device (31b) generates at least one additional instruction (Scom2) for controlling the motor vehicle, said additional image (Im2) resulting from at least one geometric and/or radiometric transformation performed on said captured image (Im1, and a digital merging module (33) configured to generate a resulting control instruction (Scom) based on said control instruction (Scom1) and said at least one additional control instruction (Scom2).