Autonomous Vehicle Object Recognition Training Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for object recognition and distance estimation in autonomous vehicles are limited due to the infinite variations of road objects, processing capabilities, and the limitations of cameras, radar, and LiDAR technologies, leading to inaccuracies and high costs.
Innovation Solution
A method involving the use of multiple recognition and detection techniques, such as YoloV4-CSP and YoloV4-P7 algorithms, to generate improved training data videos by integrating and sampling frames, and calculating transform values for object coordinates across frames, enhancing object recognition and distance estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple recognition techniques are applied to improve object recognition accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the object recognition task into multiple independent recognition techniques (e.g., YoloV4-CSP, YoloV4-P7) that process images separately. Each technique operates as an independent module, allowing the system to leverage multiple algorithms without creating a monolithic complex system. The segmentation enables modular processing where each recognizer handles specific aspects of object detection.
Solution Approach 2:
The patent merges the results from multiple recognition techniques through frame integration and sampling. By combining outputs from different recognition algorithms and using detection techniques to integrate frames, the system achieves improved measurement precision while managing complexity through structured combination rather than chaotic integration.
2Reliability
If more frames are integrated to improve recognition reliability, then reliability is improved, but loss of time increases
Solution Approach 1:
The patent applies partial action by using sampling techniques to select only necessary frames from the integrated frame set. Instead of processing all possible frames, the system extracts a representative subset that maintains recognition reliability while significantly reducing processing time. This partial processing approach avoids the excessive time cost of complete frame analysis.
Solution Approach 2:
The patent performs preliminary frame integration and detection before final sampling. By pre-processing frames to identify and integrate relevant detection results beforehand, the system prepares data in advance, allowing faster final processing. This preliminary action reduces the time burden during critical recognition phases.
3Manufacturing precision
If advanced recognition algorithms are used to improve measurement precision, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The patent employs universal recognition algorithms like YoloV4-CSP and YoloV4-P7 that can handle multiple object types and scenarios through a single framework. These multi-functional algorithms provide high manufacturing precision for distance estimation without requiring separate specialized systems for each object type, thereby managing algorithmic complexity while maintaining accuracy.
Data Source
AI summary
A method of generating an improved training data video includes recognizing an object included in a first image by applying at least two recognition techniques to the first image acquired during driving, applying at least two detection techniques to a result of recognizing the object and detecting a frame by each of the applied detection techniques, generating a frame set including a plurality of frames by integrating the detected frames, and generating a second image by sampling the integrated frame set.