Multi-Plate Recognition via Edge Neural Network Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic license plate recognition (ANPR) systems face challenges such as poor file resolution, blurry images, poor lighting, object obstruction, non-standard fonts, and limited real-time tracking capabilities due to reliance on single frame analysis and cloud-based systems, which are hindered by network performance issues.
Innovation Solution
The implementation of a camera system that continually captures video and uses a vision processor to execute a neural network model for real-time license plate detection and recognition, aggregating frames for improved character detection and optical character recognition (OCR), and employing specialized circuitry to optimize CPU usage and handle multiline and stacked character recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If single frame analysis methods are used for license plate detection, then the system complexity is reduced, but the real-time tracking capability and measurement precision are limited
Solution Approach 1:
The patent implements continuous video frame capture and processing instead of single frame analysis. The system continuously monitors video streams, detecting license plates across multiple consecutive frames to maintain persistent tracking. This continuous action improves detection precision by allowing the system to accumulate evidence across frames while maintaining real-time capability through efficient processing pipelines.
Solution Approach 2:
The system performs preliminary processing on video frames by detecting motion and identifying potential license plate regions before full OCR analysis. This preliminary detection stage filters out non-relevant frames and pre-locates license plate areas, reducing the computational burden on subsequent recognition stages while improving overall detection precision.
2Power
If cloud-based systems are used for license plate recognition, then computational power is increased, but network performance issues limit real-time processing speed
Solution Approach 1:
The patent implements edge computing by deploying neural network models directly on local devices such as cameras, gate controllers, or on-premise servers. This local processing eliminates network dependency for real-time recognition while maintaining high computational power through specialized hardware like GPUs or NPUs. The system processes video frames locally to generate license plate readings without requiring continuous cloud connectivity.
Solution Approach 2:
The recognition system is segmented into modular components that can be distributed across local and cloud environments. Critical real-time functions like detection and basic OCR are performed locally, while non-critical functions like database queries or archival storage can utilize cloud resources. This segmentation allows the system to maintain real-time processing speed for essential functions while still leveraging cloud computational power when available.
3Measurement precision
If multiple video frames are aggregated for license plate recognition, then measurement precision is improved, but the processing time and device complexity increase
Solution Approach 1:
The system aggregates only a limited number of consecutive video frames (e.g., 3-5 frames) rather than processing entire video sequences. This partial action provides sufficient temporal redundancy to improve character detection accuracy through frame aggregation while capping the processing time to prevent excessive computational burden. The system stops aggregation after a predetermined frame count is reached.
Solution Approach 2:
The system performs frame aggregation at periodic intervals rather than continuously processing every frame. Motion-triggered detection activates aggregation only when vehicle presence is detected, and the system processes frames in periodic batches rather than continuously. This periodic approach reduces overall processing time while maintaining detection accuracy through strategic frame sampling.
Data Source
AI summary
A method and related system operations include obtaining a video stream with an image sensor of a camera device, detecting a plurality of target objects by executing a neural network model based on the video stream with a vision processor unit of the camera device. The method also includes generating a plurality of bounding boxes, determining a plurality of character sequences by, for each respective bounding box of the plurality of bounding boxes, performing a set of optical character recognition (OCR) operations to determine a respective character sequence of the plurality of character sequences. The method also includes updating a plurality of tracklets to indicate the plurality of bounding boxes and storing the plurality of tracklets in association with the plurality of character sequences in a memory of the camera device.