M6 Project: Video Surveillance for Road Traffic Monitoring

Group 06

Team members:

• Daniel Fuentes - [email protected] - DanielFuentes16
• Daniel Yuste - [email protected] - DanielYG
• Sergi García Sarroca - [email protected] - SoftonCV
• Isaac Pérez Sanz - ipesanz

Project Description:

This project is related to basic concepts and techniques related to video sequences mainly for surveillance applications. It contains the work of the authors along 5 weeks achieving different objectives:

• The use of statistical models to estimate the background information of the video sequence.
• Use of deep learning techniques to detect the foreground.
• Use optical flow estimations and compensations.
• Track detections.
• Analyze system performance evaluation.

Week 1: Evaluation Metrics:

Main Tasks:

(a) Detection metrics:

• [] IoU & mAP for (ground truth + noise)
• [] mAP for provided object detections

(b) Detection metrics. Temporal analysis:

• [] IoU vs time

(c) Optical flow evaluation metrics:

• MSEN: Mean Square Error in Non-occluded areas
• PEPN: Percentage of Erroneous Pixels in Non-occluded areas
• Analysis & Visualizations

(d) Visual representation optical flow:

• Optical Flow Plot

Visual Results:

Optical Flow

Optical Flow Plot

Week 2: Bakckground Substraction:

Main Tasks:

Update Missing Work from Week 1 :

(a) Detection metrics:

• IoU & mAP for (ground truth + noise)
• mAP for provided object detections

(b) Detection metrics. Temporal analysis:

• IoU vs time

Week 2 tasks:

(a) Background Estimation:

• Gaussian Modelling
• Evaluation

(b) Stauffer & Grimson:

• Adaptive Modelling
• Comparison with task (a)

(c) Comparison with state-of-the-art

(d) Adding Color Spaces

Visual Results:

Background Estimation

Denoised Test

Week 3: Segmentation, Object Detection & Tracking:

Main Tasks:

(a) Object detection:

• Off-the-Shelf
• Fine-tune to your data

(b) Object tracking

Visual Results:

Fast R-CNN

Mask R-CNN

Week 4: OpticalFlow & Tracking:

(a) Optical Flow:

• Optical Flow with Block Matching
• Off-the-Shelf Optical Flow

(b) Video stabilization:

• Video stabilization with Block Matching
• Off-the-shelf Stabilization

(c) Object Tracking:

• Object Tracking with Optical Flow

Visual Results:

Optical Flow

Video Stabilization

Week 5: Single-Camera & Multi-Camera Tracking:

Main tasks:

(a) Multi-target single-camera (MTSC) tracking:

• Read the Data & Evaluation description for Track 3 (Multiple-camera tracking).
• Obtain results for SEQ 1 & SEQ 4

(b) Multi-target multi-camera (MTMC) tracking

Visual Results:

S03 - c010 & c011

S04 - c038 & c040

• Slides for the project: T06-Google Slides

• Link to the article: Group06-Report