Real-life-Anomaly-Detection

This project under SoC, deploys an Anomaly Algorithm (Unsupervised) to detect distinct pedestrian behaviour in surveillance footages.

What All I Have Learned

• CNN and Its Implementation

  • Learned how convolution, padding, and strided convolution work. Also explored notations in CNN.

• Filters and Pooling

  • Learned to apply filters and pooling layers along the RGB channels.

• OpenCV

  • Learned the basics of image manipulation using the OpenCV library and working on webcam and video as well.

• YOLOv5

  • Learned sliding window detection and its disadvantages.
  • Learned the concept of overlapping and I/O ratio.
  • Learned anchor boxes.
  • Learned the YOLOv5 mechanism.

• Implementing YOLOv5 on MOT17 Pedestrian Dataset

  • Implemented YOLOv5 model (small, medium, and medium with frozen layers) on the dataset.
  • Used YOLOv5 trained model and basic anomaly detection techniques to observe anomalies across the test avenues data provided by CUHK.