Visual Perception Network for Local Motion and Pose Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual perception technologies have coarse perception granularity, making it impossible to analyze and understand local motion of objects, which is critical for applications like autonomous driving where responding to specific object movements is necessary.
Innovation Solution
A visual perception method using a perception network that recognizes and classifies moving components of objects, such as vehicle doors or human limbs, to determine their pose and generate control commands for appropriate responses, combined with a perception network training method that uses 3D models to generate edited images for training, refining perception granularity and improving analysis of local motion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bounding box and contour-based shallow understanding is used, then the perception system can identify the whole target object, but it cannot analyze local motion of the object
Solution Approach 1:
The patent segments the perception task into two levels: (1) whole object detection using bounding boxes and contours, and (2) local component detection by dividing the object into multiple movable components (e.g., doors, windows, trunk). This segmentation allows the system to analyze local motion while maintaining the simplicity of whole-object detection.
Solution Approach 2:
The patent applies local quality by treating different parts of the object differently - the whole object is detected using standard bounding box methods, while specific movable components (doors, windows, trunk) are detected using specialized local detection algorithms. This allows the system to focus computational resources on parts that require detailed motion analysis.
2Reliability
If the perception system only detects the whole target object, then the system complexity remains low, but the system cannot respond to specific local motions for safety-critical decisions
Solution Approach 1:
The patent segments the object into movable components (doors, windows, trunk) and detects their individual states. This segmentation enables the system to make reliable safety decisions based on local motion (e.g., detecting an open door) without requiring a complete redesign of the perception system.
Solution Approach 2:
The patent performs preliminary classification to identify which objects have movable components, then applies local detection algorithms only to those objects. This preliminary action reduces unnecessary computational complexity while ensuring reliable detection when needed.
3Measurement precision
If fine-grained component-level detection is implemented, then local motion analysis becomes possible, but the detection and measurement difficulty increases
Solution Approach 1:
The patent applies local quality by using different detection strategies for different components. Whole objects use standard bounding box detection, while movable components (doors, windows, trunk) use specialized local detection algorithms that consider their specific motion characteristics and spatial relationships to the main object.
Solution Approach 2:
The patent uses the bounding box and contour detection results as an intermediary step before performing component-level detection. The detected bounding box provides a search region that simplifies subsequent component detection, and the contour information serves as a reference for identifying movable components.
Data Source
AI summary
The present disclosure provides a visual perception method and apparatus, a perception network training method and apparatus, a device and a storage medium. The visual perception method recognizes the acquired image to be perceived with a perception network to determine a perceived target and a pose of the perceived target, and finally determines a control command according to a preset control algorithm and the pose, so as to enable an object to be controlled to determine a processing strategy for the perceived target according to the control command. According to the perception network training method, acquire image data and model data, then generate an edited image with a preset editing algorithm according to a 2D image and a 3D model, and finally train the perception network to be trained according to the edited image and the label.