LiDAR Object Recognition Pipeline for Faster Point Cloud Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition devices using LiDAR sensors face high processing loads and long processing times due to the large amount of point cloud data, leading to increased manufacturing costs and reduced recognition speed for sudden changes in the environment.
Innovation Solution
An object recognition device with a software pipeline architecture that includes multiple processing units and task queue regions, where output data is rejected if invalid, and a task priority decision unit dynamically allocates resources to process valid data in parallel.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If point cloud data from LiDAR sensor is processed using conventional methods, then recognition accuracy can be maintained, but processing load increases and processing time becomes long
Solution Approach 1:
The patent divides the point cloud data processing into multiple stages: data acquisition, preprocessing (filtering, downsampling), feature extraction, and object recognition. Each stage processes data independently and passes results to the next stage, enabling parallel processing and reducing overall processing time while maintaining recognition accuracy.
Solution Approach 2:
The patent performs preliminary processing operations such as filtering out invalid points and downsampling the point cloud data before main recognition processing. This preliminary action reduces the data volume that requires intensive processing, thereby decreasing processing load and time without compromising final recognition accuracy.
2Productivity
If high-performance computer is used to handle large processing load, then processing speed improves, but manufacturing cost increases
Solution Approach 1:
The patent implements dynamic resource allocation where processing resources are allocated based on the actual complexity and volume of incoming point cloud data. The system can scale processing power usage dynamically, using more resources when needed and conserving resources when data is simple, thereby improving processing speed without requiring constantly high-performance hardware.
Solution Approach 2:
The patent changes processing parameters such as downsampling rate, filtering thresholds, and feature extraction depth based on data characteristics and processing queue status. By dynamically adjusting these parameters, the system optimizes processing speed for different scenarios without requiring maximum hardware performance continuously, reducing manufacturing costs.
3Loss of time
If distributed processing is implemented across multiple devices, then processing time shortens, but system complexity increases
Solution Approach 1:
The patent combines multiple processing functions (filtering, downsampling, feature extraction, recognition) into an integrated software pipeline that runs on a single processing system. This merging approach enables parallel processing of different data streams and stages without the communication overhead and synchronization complexity of distributed hardware systems, reducing processing time while controlling system complexity.
Data Source
AI summary
An object recognition device (3) includes: a processor (4) that inputs point cloud sensor data sensed by a sensor (externality recognition sensor (1)) to a software pipeline formed of a plurality of stages and executes object recognition processing; and a memory. The processor (4) has a plurality of processing units (5) that executes processing constituting the object recognition processing allocated to each of stages A to C. The memory has a plurality of task queue regions (task queues A to C) that temporarily stores the output data of each of the stages. In a case where the output data of each of the stages is not valid upon input to the subsequent stage, the processor (4) rejects the output data without storing same in the respective task queue regions.