Laser Rangefinder Object Classification via Point Cloud Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies are inadequate for accurately detecting, classifying, and estimating the properties of moving vehicles, such as position, speed, heading, and type, in a dynamic road environment, particularly in distinguishing between different types of vehicles like motorcycles, cars, and trucks.
Innovation Solution
A method using scanning laser rangefinders that groups point clouds into a common reference frame, determines sets of points, creates bounding boxes, segments the shape of objects, and employs Kalman filtering and belief theory for tracking and classification, updating tracks based on classification results to refine movement estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If scanning laser rangefinders are used to detect moving objects, then measurement capability is improved, but classification precision between different vehicle types deteriorates
Solution Approach 1:
The patent segments the detection process into multiple stages: initial detection using laser rangefinders, extraction of geometric features (bounding boxes, segments), classification based on feature patterns, and refined tracking. This segmentation allows the system to leverage the strengths of laser detection while adding specialized classification capabilities through geometric analysis of detected objects.
2Device complexity
If simple detection methods are used, then device complexity is reduced, but classification capability deteriorates
Solution Approach 1:
The patent creates a multi-functional system where the laser rangefinder serves both detection and measurement functions, the geometric feature extraction serves both characterization and classification functions, and the tracking system serves both motion estimation and classification refinement functions. This universal approach allows a single integrated system to handle multiple tasks without requiring separate specialized devices for each function.
3Loss of information
If detailed classification is implemented, then object identification is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary geometric feature extraction (bounding boxes, segments) immediately upon detection, before full classification is required. These pre-computed features are then reused across multiple classification and tracking operations, avoiding redundant calculations and reducing overall processing time while maintaining detailed classification capability.
4Reliability
If tracking is updated continuously, then temporal accuracy is improved, but computational load increases
Solution Approach 1:
The patent implements a feedback mechanism where classification results from previous time steps inform the tracking updates in current time steps. The system uses predicted trajectories and classification confidence to selectively update track parameters, reducing computational load by avoiding unnecessary recalculations while maintaining temporal accuracy through iterative refinement based on new sensor data.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables precise detection and classification of vehicles, including type, speed, and heading, with improved temporal tracking of their movement, effectively distinguishing between various vehicle types.
Implementation Method 1
The scanning laser rangefinder, LiDAR (English acronym for 'Light Detection and Ranging'), is the only detector currently capable of directly measuring this information.
Implementation Method 2
laser impacts on at least one object to be classified
Data Source
AI summary
A method for classifying at least one object to be classified around a motor vehicle comprising at least one laser range-finder, comprising the following steps: acquiring a point cloud comprising at least two points coming from laser impacts on at least one object to be classified, grouping the point clouds, determining sets of points among the point clouds that have been grouped according to their respective coordinates and a maximum distance between points, producing a prediction of a path characterising the presence and movement of the object over time by determining the state of the paths at the current iteration on the basis of the state of the paths at the preceding iteration and the predictive equations of the Kalman filter, determining association information of targets and paths on the basis of a bounding box associated with each target and with each path, determining a classification of the objects to be classified among a plurality of categories on the basis of lists of segments and values of criteria in order to determine the type of vehicle corresponding to each object.