Overview

Validation gives insight on how well a model or software performs. In the case of OpenTrafficCam, we want to evaluate and compare different object detection models on how well they perform in object detection and object tracking. Furthermore, we want to analyse the results of OTAnalytics on a set of traffic performance metrics.

For this purpose we are currently developing OTValidate which allows the user to analyse and compare the models' performances on object detection and tracking by calculating metrics corresponding to the specific task at hand. As mentioned above OTValidate will also provide tools to analyse the results of OTAnalytics in regards to traffic performance.

OTValidate

The image below gives a good overview on the structure of OTValidate:

For the object detection task OTValidate needs two input files namely an .otdet file and the ground truth label data for the object detection task. Alternatively, a custom or an existing model can be given over as an input instead of an .otdet file.