Compressed Random Forest Obstacle Classification for Unmanned Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing obstacle classification systems for unmanned vehicles require significant memory and space due to the storage of extensive information in random forest models, leading to slow and inefficient classification processes.
Innovation Solution
Implementing a compressed random forest model with decision trees that store only feature index, classification threshold, and node position information in non-leaf nodes, and class index information in leaf nodes, reducing the amount of data stored and processed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a random forest model with multiple tree structures and nodes storing multiple fields is used for obstacle classification, then the classification accuracy is improved, but the memory occupancy and space requirements increase significantly
Solution Approach 1:
The patent extracts and removes unnecessary information fields from the random forest model nodes. Specifically, it eliminates redundant feature values, class labels, and other non-essential data that were originally stored in each node, keeping only the critical structure information (node position, feature index, threshold, and split direction). This extraction principle directly reduces memory occupancy while preserving the model's classification functionality.
Solution Approach 2:
Instead of storing complete information in each node as in traditional random forest models, the patent inverts the approach by storing only minimal structural information and reconstructing the full context during traversal. The model stores compressed representations (node position indices, feature indices, thresholds) and reconstructs the decision logic on-the-fly during classification, thereby reducing storage requirements while maintaining accuracy.
2Reliability
If extensive information is stored in the nodes of the random forest model, then the classification completeness is improved, but the classification speed decreases
Solution Approach 1:
The patent extracts only the essential elements needed for classification from the node information. By removing redundant data storage and keeping only critical parameters (feature index, threshold, split direction, node position), the model reduces the amount of information that needs to be processed during classification, thereby increasing classification speed while maintaining completeness through the preserved decision logic.
Solution Approach 2:
The patent changes the parameter representation in the model from storing complete information to storing compressed parameters. Each node stores minimal parameters (position index, feature index, threshold value, split direction) instead of full information, and these parameters are efficiently processed during traversal to achieve faster classification while maintaining decision accuracy.
3Quantity of substance
If a compressed random forest model storing only essential information is used, then the memory efficiency is improved, but the system complexity increases
Solution Approach 1:
The patent segments the information storage into distinct, well-defined components: node position index, feature index, threshold value, and split direction. This segmentation organizes the compressed data structure into manageable, clearly-defined fields, making the system easier to implement and maintain despite the compression. Each segment serves a specific function, reducing overall system complexity through modular organization.
Data Source
AI summary
An obstacle classification method and apparatus based on unmanned vehicle, a device, and a storage medium are provided. The method includes: obtaining obstacle information of a plurality of obstacles detected by a detection device of the unmanned vehicle; and performing classification processing on the obstacle information using a compressed random forest model, to obtain a result of obstacle classification, where a non-leaf node of each decision tree of the compressed random forest model stores only feature index information, classification threshold index information and node position index information, the node position index information is left or right node position index information; a leaf node of each decision tree stores only class index information. Only part of information needs to be stored, and there is no need to occupy more memory and space of the unmanned vehicle system, thereby reducing the memory occupancy, and improving the speed and efficiency of classification.