3D Object Classification Through Body-Taper Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D sensing technologies, such as LiDAR, struggle with accurate classification of objects like humans due to false positives from volumetric analysis, especially in crowded spaces and adverse weather conditions, leading to unreliable object classification.
Innovation Solution
A system and method using body taper detection, calculating a unique taper ratio through horizontal layers from 3D sensors, combined with volume analysis, to classify objects by generating best-fit curves with golden ratios for real-time classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If volumetric analysis is used to classify objects, then classification can be performed, but false classifications occur especially in crowded spaces and adverse weather conditions
Solution Approach 1:
The patent segments the object detection process into multiple independent analysis components: volumetric analysis, taper ratio analysis, and horizontal layer analysis. Each component evaluates different geometric properties of detected objects, and their results are combined to make the final classification decision. This segmentation allows the system to maintain high processing speed while improving accuracy by cross-validating multiple features.
Solution Approach 2:
The patent introduces new geometric parameters (taper ratio and horizontal layer dimensions) to complement the traditional volumetric analysis. By changing from a single parameter (volume) to multiple parameters (volume, taper ratio, layer dimensions), the system achieves more reliable classification without sacrificing processing efficiency, as each parameter provides complementary information about object shape and structure.
2Device complexity
If only volumetric analysis is used, then processing is simple and fast, but objects of similar volume but different shapes cannot be distinguished
Solution Approach 1:
The patent transitions from one-dimensional volumetric analysis to multi-dimensional geometric analysis by introducing taper ratio (ratio of horizontal to vertical dimensions) and horizontal layer dimension analysis. This dimensional expansion allows the system to distinguish between objects of similar volume but different shapes, such as differentiating human bodies from animals or inanimate objects, while maintaining computational efficiency through standardized calculation methods.
3Reliability
If complex classification algorithms are used to improve accuracy, then false classifications are reduced, but processing time and computational resources increase
Solution Approach 1:
The patent segments the classification task into independent geometric feature extractions (volume, taper ratio, horizontal layers) that can be computed in parallel. Each feature is calculated using straightforward geometric formulas rather than complex iterative algorithms, and the final classification decision is made by comparing these features against predefined thresholds or patterns, achieving high accuracy with minimal processing time.
Solution Approach 2:
The patent replaces complex machine learning classification algorithms with a geometric feature-based classification system. Instead of using resource-intensive neural networks or support vector machines, the system uses analytically computable geometric parameters (volume, taper ratio, layer dimensions) and compares them against known patterns for different object types, achieving real-time classification with minimal computational overhead.
Data Source
AI summary
A method is provided for classification of objects by body taper detection. The method includes receiving point cloud data of a plurality of objects from one or more 3D sensors disposed in a space to be monitored, pre-processing the fused point cloud data to search, detect and segment the plurality of objects, each segmented object comprising a set of horizontal layers of interception from the 3D sensor representing the structure of the segmented object, calculating an inference ratio of each segmented object using lengths of respective horizontal layers, distances between set of horizontal layers and a distance of each segmented object from the one or more 3D sensors, checking whether the inference ratio of the segmented object lies on best fit curves pre-generated using a plurality of golden ratios associated with a particular class of objects, and accordingly determining the class of each of the segmented object.