Autonomous Vehicle Vision Control With Adaptive Pixel Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in efficiently processing sensory data from various sensors due to the computational demands of Deep Neural Networks (DNNs), which require large training datasets and are less adaptable to varying operational environments, leading to performance deterioration when environments differ from the training scope.
Innovation Solution
A computer device equipped with a low-resolution recurrent neural network that pre-processes and post-processes sensor data, allowing it to adaptively select inputs and perform tasks like image classification, road segmentation, and object detection with reduced computational load, using a System on a Chip (SoC) for independent operation without external network connectivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Deep Neural Networks (DNNs) are used to process sensory data, then classification accuracy is improved, but computational load increases
Solution Approach 1:
The patent segments the image processing task by dividing the image into multiple patches and processing them separately through the neural network. This allows the system to achieve accurate classification while reducing the computational load on any single processing unit, as each patch requires fewer calculations than processing the entire high-resolution image at once.
Solution Approach 2:
The patent transforms the spatial dimension of image processing by converting 2D image patches into 1D feature vectors through flattening and pooling operations. This dimensional transformation reduces the complexity of the data structure while preserving essential features, enabling accurate classification with reduced computational requirements.
2Measurement precision
If Deep Neural Networks (DNNs) are trained on large datasets, then performance is improved, but training time and data requirements increase
Solution Approach 1:
The patent applies partial action by using a simplified version of DNN processing that focuses on extracting essential features from image patches rather than performing exhaustive analysis. The use of pooling operations and feature selection techniques allows the system to achieve sufficient accuracy with reduced training data and shorter training time, without requiring complete processing of all possible image variations.
3Measurement precision
If high-resolution images are processed, then feature identification accuracy is improved, but processing speed decreases
Solution Approach 1:
The patent segments high-resolution images into smaller patches that can be processed more quickly. By dividing the image into manageable sections and processing them in parallel or sequence through the neural network, the system maintains feature identification accuracy while significantly improving processing speed compared to analyzing the entire high-resolution image at once.
Solution Approach 2:
The patent extracts essential features from image patches using pooling operations and feature selection techniques. Instead of processing all pixels in high-resolution images, the system extracts and retains only the most relevant features, thereby maintaining identification accuracy while reducing the computational burden and increasing processing speed.
4Device complexity
If passive vision systems are used, then system simplicity is maintained, but adaptability to varying environments deteriorates
Solution Approach 1:
The patent introduces dynamic elements into the vision system by implementing active vision mechanisms where the neural network adaptively adjusts its processing based on the input image characteristics. The system dynamically selects which patches to process, adjusts pooling strategies, and modifies feature extraction parameters according to the specific environmental conditions, thereby improving adaptability while maintaining relative system simplicity.
Data Source
AI summary
A computer device and system for controlling an autonomous vehicle are provided. The computer device comprises a memory and a processor, the computer device configured to be fitted to a vehicle and to communicate with a camera or sensor, the processor being configured to: pre-process an original image from the camera or sensor data from the sensor to produce an input image; present the input image to a neural network stored in the memory; wherein the neural network is trained to classify a feature in an image presented to it, the neural network having an input layer, a hidden layer and an output layer, the output layer including three outputs: a first feedback output for selecting pixels from the input image to input at the input layer at each iteration of the neural network; a second feedback output for selecting a colour channel of the selected pixels to input at the input layer at each iteration; and a third output for outputting an output value indicative of a classification result from the neural network; the processor further configured to obtain the output value from the neural network; and post-process the output value from the neural network to identify a feature of the environment of a vehicle.