Traffic Detection Project

This project consists of two sub-projects:

   1. Traffic Detection based on traffic images
   2. Traffic Anamoly Detection based on traffic videos

1. Traffic Detection

Dataset :

Traffic Detection Dataset: link

Approach :

In this project we will Benchmark, Compare and Analyze some of the most recent and State of the Art Object Detection models to be able to choose what suits best for our Traffic Detection Dataset. Following are the Object Detection models we will explore:

1. YOLOv8
2. YOLOv9
3. YOLOv10

You can find the trained model weights here:
model weights

Results :

2. Anamoly Detection in Traffic Videos

Dataset :

Detection of Traffic Anomaly : DoTA

Approach :

We will use the pretrained MOVAD model and evaluate the performance on 20 video.

MOVAD is built on two primary modules: a Short-Term Memory Module, which extracts information about ongoing actions using a Video Swin Transformer (VST), and a Long-Term Memory Module integrated into the classifier, leveraging a Long-Short Term Memory (LSTM) network to consider past information and action context.

Example detection video:

References:

MEMORY-AUGMENTED ONLINE VIDEO ANOMALY DETECTION
MOVAD repo
YOLOv10: Real-Time End-to-End Object Detection
yolov10-vs-yolov9