Point Cloud Classification With Ghost Object Detection for ADAS
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Active detection and ranging systems, such as radar, suffer from the issue of detecting 'ghost objects' due to signal reflections from secondary objects, which can lead to incorrect object classification and unsafe actions in advanced driver-assistance systems (ADAS).
Innovation Solution
A computer-implemented method using a machine learning model with a shared encoder and separate object and ghost classification networks to classify points in a point cloud, employing latent representations and multi-task loss functions for improved accuracy and differentiation between real and ghost objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single classification network is used to classify objects in the point cloud, then the device complexity is reduced, but the classification precision for ghost objects deteriorates
Solution Approach 1:
The classification network is segmented into two separate specialized networks: an object classification network for identifying real objects and a ghost classification network for detecting ghost objects. Each network is trained independently on specific datasets, allowing them to specialize in their respective tasks and achieve higher precision without increasing overall system complexity.
Solution Approach 2:
The system employs a shared encoder network that serves multiple functions by extracting features for both object and ghost classification tasks simultaneously. This multi-functional component reduces redundancy while maintaining the ability to perform both classification functions with high accuracy.
2Measurement precision
If separate classification networks are used for object and ghost classification, then the classification precision is improved, but the device complexity increases
Solution Approach 1:
The encoder networks from both the object classification system and ghost classification system are merged into a single shared encoder. This consolidation reduces the total number of parameters and computational resources required, thereby decreasing device complexity while preserving the precision benefits of separate classification heads.
3Productivity
If traditional machine learning classification is used for point cloud data, then the processing speed is adequate, but the classification accuracy for ghost objects deteriorates
Solution Approach 1:
The system transforms the classification approach by changing key parameters: using point cloud data with rich spatial and feature information instead of traditional structured input, and employing deep learning architectures with multiple layers of feature extraction. These parameter changes enable the system to achieve high ghost object detection accuracy while maintaining efficient processing speeds through optimized network architectures.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method (200) for classifying points in a point cloud (220) obtained by an active detection and ranging system is provided. The method comprises: computing a latent representation of the point cloud, determining from the latent representation, for each point in the point cloud, a probability that the point represents an object belonging to one or more object classes, and determining from the latent representation, for each point in the point cloud, a probability the point represents a ghost object. An object classification network (203) is used for determining the probability that the point represents an object, and a ghost classification network (205) is used for determining the probability the point represents a ghost object.