Challenge
A major technology leader faced significant challenges in deploying AI applications on embedded Edge devices, often encountering sub-optimal performance issues. They needed to demonstrate their advanced AI capabilities through multiple real-world applications spanning smart city, retail, security, and surveillance sectors. The project required developing six distinct AI applications while ensuring optimal performance on Edge devices.
Solution
Quest Global engineered a suite of AI applications on the Renesas RZ/V2L platform:
Developed an advanced footfall counting system using YOLOv3-tiny and tracking algorithms for real-time monitoring and precise counting within designated zones
Implemented a smart parking solution with custom-trained models on extensive labelled datasets, enabling accurate empty slot detection and real-time availability updates
Created a sophisticated face authentication system utilizing FaceNet technology for facial landmark detection and ID verification, streamlining security processes
Engineered an expiry date detection system combining YOLOv3-tiny and Tesseract OCR technology for accurate date extraction and verification
Designed a crack identification system using UNet with ResNet34 backbone for precise surface crack segmentation across various materials
Built a versatile object counting application using YOLOv3-tiny trained on custom datasets of animals, vehicles, and COCO datasets
Results – At A Glance
Achieved exceptional accuracy rates: 99% for smart parking and 99.63% for face authentication
Delivered real-time performance with 10 FPS for crack segmentation analysis
Attained 83% accuracy in expiry date detection
Demonstrated consistent object detection performance with mAP 33.1 for both footfall counting and object counting applications
