Benchmark

Version 1.0

1. Training

Model
Description
Vehicle Detection Model
Recognize vehicle bounding box on an input image.
Also classify vehicle type, color, and vehicle brand.
Vehicle License-plate Detection Model
Recognize license-plate bounding box inside on a vehicle image.
Also classify the type of license-plate.
License-plate Character Recognition Model
Recognize the license plate number/character and a registered province.

2. Model Evaluation based on iApp-vehicle image data (118 images)

Model
Accuracy (%)
Vehicle Detection Model
96.4
Vehicle License-plate Detection Model
95.65
License-plate Character Recognition Model
94.82
Overall Accuracy
95.62

3. Evaluation (video)

Test Data
Test Type
Number of Frames
Process Time per Frame
Accuracy
Video 1
Vehicles are closer
to the camera
9,397
0.04 - 0.07 seconds
95.5 %
Video 2
Vehicles are bit far from
the camera
9,003
0.04 - 0.07 seconds
82.76 %
Please note that this version only showcases the recognition of vehicle features:
  • Vehicle type.
  • Vehicle brand.
  • Vehicle color.
  • Province that the license plate is registered.
The Accuracy of these parameters will be improved and released officially in the upcoming version.
Last modified 2mo ago
Copy link