Video Surveillance for Road Traffic Monitoring

Master in Computer Vision - M6 Video analysis - Team 8

Goal: The goal of this project is to learn the basic concepts and techniques related to video sequences processing, mainly for surveillance applications. We will focus on video sequences from outdoor scenarios, with the application of traffic monitoring in mind. The main techniques of video processing will be applied in the context of video surveillance: background modelling, moving object segmentation, motion estimation and compensation and video object tracking.

Team 8

Members	Contact
Yu Pang	yu.pang@e-campus.uab.cat
Khanh Nguyen	khanhnguyen21006@gmail.com
Siddhant Bhambri	siddhantbhambri@gmail.com
Dhananjay Nahata	nahatadhananjay33@gmail.com

Explanation

We are experimenting on Python3.7

• datasets/: AICity_data dataset path, frames of video
• annotations/: useful annotations for evaluate the resluts
• utils/: useful tools to do different things, such as transfer frames to gif image

Week1

• Task 1: Segmentation metrics. Understand precision & recall.
• Task 2: Segmentation metrics. Temporal analysis.
• Task 3: Optical flow evaluation metrics.
• Task 4: De-synchornized results.

Week2

• Task 1: Gaussian distribution
  • Task 1.2&1.3: Evaluate results
• Task 2: Adaptive modelling.
  • Task 2.1: Recursive Gaussian modeling
  • Task 2.2: Evaluate and compare to non-recursive
• Task 3: Compare with state-of-the-art
• Task 4: Color sequences

Week3

• Task 1: Object detection

  • Task 1.1: Off-the-shelf
  • Task 1.2: Fine-tune to your data
  • Task 1.3: K-Fold Cross-validation

• Task 2: Object tracking

  • Task 2.1: Tracking by overlap
  • Task 2.2: Tracking with a Kalman Filter
  • Task 2.3: IDF1 score

Week4

• Task 1: Optical Flow

  • Task 1.1: Optical Flow with Block Matching
  • Task 1.2: Off-the-shelf Optical Flow

• Task 2: Video stabilization

  • Task 2.1: Video stabilization with Block Matching
  • Task 2.2: Off-the-shelf Stabilization

• Task 3: Object Tracking

  • Task 3.1: Object Tracking with Optical Flow
  • Task 3.2: (optional) CVPR 2021 AI City Challenge

Week5

• Task 1: Multi-target single-camera (MTSC) tracking
• Task 2: Multi-target multi-camera (MTMC) tracking

Documentation

• Final presentation slides
• Final report overleaf