In this Project, we compare a few object detection models, when applied to Car Detection.
This process is done on several Maltese Video-feeds, where non-deal conisiotns are met, such as heavy rain, sun shining in the camera and video compression.
We have labelled our own datatset based on different parts of these same video feeds, where the Object Detection models where trained to these using Tranafser Learning.
This has shown that the results indeed improved in most cases, where they were empircally comapres through error metrics.
Finally, A centroid- based system was adopted for car counting, where individuals cars were counted as well as the carriageway they were moving towards.