CodeSense
- #Facilities
A real-time solution featuring a two-step detection system capable of identifying barcodes through object detection and item code numbers using Optical Character Recognition technology.
- Computer Vision
- Machine Learning
Impact
Our software provides significant business value by automating parcel reception, reducing processing time and improving efficiency. The accuracy of recognition for barcodes is 97%. This solution can benefit various industries worldwide that rely on shipment services.
- Automating parcel reception minimises the likelihood of errors and improves the overall quality of service.
- Additionally, it can streamline operations and result in cost savings for businesses.
Services we provided
A barcode reader plug-in for NX Witness software
Tech Stack
PaddleSpeech
C++
Cmake
AWS
Flask
Challenges and Solutions
🧐 Challenges
- Object detection and OCR accuracy: Overcoming the challenge of achieving high accuracy through extensive research and model optimisation.
- Model size optimisation: Reducing model size without sacrificing accuracy through quantisation techniques.
- Software deployment complexity: Deploying the software in C++ using CMake requires significant effort for efficient performance and compatibility.
- Compatibility testing and optimisation: Ensuring efficient performance on diverse hardware configurations through thorough testing and optimisation
💡 Solutions
During our work, we accomplished the following:
- trained our model using the PaddlePaddle framework and evaluated its results on an independent dataset.
- to optimise model size, we employed quantisation.
- deployed the solution using CMake software written in C++.
User flow
1. The video stream is constantly read from the camera.
2. Every n-th frame is processed by the system.
3. A bounding box is detected for the barcode, and the text within it is extracted.
4. The results are shown in the desktop application, and the text is saved to the database.