Synthetic Image Generation for CNN Obstruction Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating training image sets for CNNs to detect rare objects in autonomous driving scenarios is challenging due to the difficulty in collecting diverse data, especially for unique objects like animals, which are not frequently encountered in normal driving conditions.
Innovation Solution
A method involving a computing device that acquires original images and synthesized labels with bounding boxes for specific objects not present in the original images, using a CNN module to generate synthesized images and labels by adding specific objects and labels at corresponding locations, and iteratively adjusting these using a random seed to create more varied and realistic training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If training images are collected from normal driving video data, then common objects like humans, bicycles, and cars can be easily acquired for CNN learning, but unique objects like tigers or alligators cannot be acquired and the learning device cannot learn parameters for these rare objects
Solution Approach 1:
The patent uses image synthesis technology to create artificial training images by copying and pasting rare objects (like tigers or alligators) onto backgrounds from normal driving videos. This allows the system to generate training data for rare objects without needing to capture them in real driving conditions, thus resolving the contradiction between ease of data collection and ability to detect rare objects
Solution Approach 2:
The system performs preliminary synthesis of training images containing rare objects before the CNN learning process. By pre-generating these synthetic training images with rare objects placed in various driving scene contexts, the system prepares the necessary training data in advance, enabling the CNN to learn detection capabilities for rare objects that would otherwise be impossible to collect
2Reliability
If a CNN is trained only on common objects from real driving data, then the model achieves good performance on frequent objects, but fails to detect rare objects that do not appear in the training data
Solution Approach 1:
The image synthesis system creates a universal training dataset that serves multiple purposes: it maintains all the common driving scene elements from original videos while adding rare objects. This multi-functional approach allows the same training images to teach the CNN both common object detection (from the original video content) and rare object detection (from the synthesized additions), eliminating the need for separate training processes
Data Source
AI summary
A method of generating at least one training data set including steps of: (a) a computing device acquiring (1) an original image and (ii) an initial synthesized label generated by using an original label and a bounding box corresponding to an arbitrary specific object (b) the computing device supporting a CNN module to generate a first synthesized image and a first synthesized label by using the original image and the initial synthesized label, wherein the first synthesized label is created by adding a specific label to the original label at a location in the original label corresponding to a location of the bounding box in the initial synthesized label, and wherein the first synthesized image is created by adding a specific image o to the original image at a location in the original image corresponding to the location of the bounding box in the initial synthesized label.