M6 Video Analysis - MCV

Video Surveillance for Road Traffic Monitoring

Team 7

Members	Mail	Github
Alex Tempelaar	alexander.tempelaar@e-campus.uab.cat	Tempe25
Víctor Ubieto	victor.ubieto@e-campus.uab.cat	victorubieto
Mar Ferrer	mar.ferrerf@e-campus.uab.cat	namarsu
Antoni Rodríquez	antoni.rodriguez@e-campus.uab.cat	antoniRodriguez

• Link to our final report in LateX here
• Presentation slides linked at each week explanation
• Dataset used: In this project we used the dataset from CVPR 2020 AI City Challenge, specifically the sequence 03 for testing, and sequences 01 and 04 for training.

Project Schedule

Week 1 (Slides)

• Task 1: Detection metrics.
• Task 2: Detection metrics. Temporal analysis.
• Task 3: Optical flow evaluation metrics.
• Task 4: Visual representation optical flow.

Process to run the code

1. Add paths to data files:
   • Task 1.1
     • Groundtruth xml path: --Lines 17
   • Task 1.2
     • Groundtruth xml path: --Lines 78
     • Prediction paths: --Lines 80-82
   • Task 2
     • Groundtruth xml path: --Line 105
     • Video path: --Line 106
     • Prediction paths: --Lines 108-110
   • Task 3
     • Flow prediction: --Line 148
     • Groundtruth flow: --Line 149
   • Task 4
     • Flow prediction: --Line 173
     • Groundtruth flow: --Line 174
     • Image path: --Line 175
2. Select the task to run leaving it uncommented: --Lines 184-188
3. Run: >> python lab1.py

Week 2

• Task 1: Gaussian modeling of the background for foreground extraction and evaluation.
• Task 2: Adaptive modeling and evaluation.
• Task 3: State-of-the-art methods for foreground extraction.
• Task 4: Use color spaces to perform foreground extraction.

Process to run the code

1. Add paths to data files:
   • Task 1.1
     • Groundtruth xml path: --Lines 11
     • Video path: --Lines 15
   • Task 1.2
     • The user variables are the same than in task 1.1
   • Task 2
     • Groundtruth xml path: --Line 51
     • Video path: --Line 55
     • Prediction paths: --Lines 108-110
   • Task 3
     • Groundtruth xml path: --Lines: 103
     • Video path: --Lines: 106
   • Task 4
     • The experiments were executed using the previous code.
2. Select the task to run leaving it uncommented: --Lines 126-128
3. Run: >> python lab2.py

Week 3

• 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

Process to run the code

1. Add paths to data files:
   • Task 1.1
     • Model name: --Line 21
     • Groundtruth xml path: --Lines 22
     • Video path: --Lines 23
   • Task 1.2_B (Cross_val technique B)
     • Hyperparameters: --Lines 72-74
     • Gt path: -- Line 78
     • Video path: --Line 79
   • Task 1.2_C (Cross_val technique C)
     • Hyperparameters: --Lines 195-197
     • Gt path: -- Line 201
     • Video path: --Line 202
   • Task 2.1
     • Pkl path (with bboxes and socores): --Lines: 316
     • Video path: --Lines: 317
     • Gt path: --Line 318
     • Parameters: --Lines 319-323
   • Task 2.2
     • Pkl path (with bboxes and socores): --Lines: 498
     • Video path: --Lines: 499
     • Gt path: --Line 500
     • Parameters: --Lines 501-502
2. Select the task to run leaving it uncommented: --Lines 626-630
3. Run: >> python lab3.py

Week 4

• Task 1.1: Block matching implementation
• Task 1.2: Off-the-shelf optical flow methods
• Task 2.1: Video stabilization with Block Matching
• Task 2.2: Off-the-shelf Stabilization
• Task 3.1: Object tracking with optical flow

Process to run the code

1. Add paths to data files:
   • Task 1.1
     • Perform grid search: --Line 24
     • Distance function for template matching: --Line 24
     • Image 1 path: --Line 29
     • Image 2 path: --Line 30
     • Flow gt path: --Line 33
   • Task 1.2
     • Algorithm: --Line 107
     • Image 1 path: --Line 109
     • Image 2 path: --Line 110
     • Flow gt path: --Line 113
   • Task 2.1
     • Video path: --Lines: 208
   • Task 2.2
     • Method: --Line 260
   • Task 3.1
     • Pkl file: --Line 323
     • Video path: --Lines: 324
     • Gt path: --Line 325
     • Parameters: --Lines 326-330
2. Select the task to run leaving it uncommented: --Lines 528-532
3. Run: >> python lab4.py

Week 5 (Slides)

• Task 1: Multi-target Single-camera Tracking
• Task 2: Multi-target Multi-camera Tracking

Process to run the code

1. Add paths to data files:
   • Task 1
     • Detections: --Line 16
     • Ground truth: --Line 17
     • Video: --Line 18
   • Task 2
     • Ground truth: --Line 335
     • Sequence 03 paths: --Line 336
     • Pickles path (optional): --Lines 375-376
2. (Optional) Tweak parameters and flags as desired:
   • Task 1: --Lines 16 and 21-35
   • Task 2: --Lines 379-382
3. Select the task to run leaving it uncommented: --Lines 496-497
4. Run: >> python lab5.py