Field Trip Prep: Your Data Collection Plan

Introduction

In lesson 28 you'll take a laptop to a real football field and collect a dataset. A film crew never turns up and "sees what happens": they bring a shot list. Today you'll write yours: where the camera goes, how many frames of each class you need, the settings for bright sunlight, who does which job, and what you'll do when things go wrong.

By the end of this lesson you'll be able to:

• Turn the data-quality lessons from lesson 10 to lesson 14 into concrete field capture decisions
• Design a camera position plan that maximises variety and reduces bias
• Set a target count for each class, and explain why the numbers differ
• Configure capture.py (blur threshold, interval, class) for outdoor sport
• Assign team roles for the field trip
• Write a 1-page Field Collection Brief that a teacher can review and approve
• Anticipate field-specific failure modes and write down what the team will do about each

Working principle

Why plan first?

Sessions 10 to 14 taught what makes data good, how to capture it, label it, split it, and how to treat the people in it. Now you turn that into a plan. Without a plan, teams capture whatever is in front of them, miss whole classes, and come back with unusable footage. A pre-approved brief prevents that.

Every rule in the brief must trace back to a lesson:

If you can't trace a rule back to a lesson, go back to that lesson before the trip.

Field Trip 1: overview

Anticipated field failure modes and what to do

Each team must address all of these in its brief.

capture.py field configuration

Open your capture.py from lesson 11 and change the constants at the top for outdoor football. Write each setting in your brief, with a reason.

# ---------------- FIELD TRIP CONFIGURATION (lesson 15 review, lesson 28 deployment) ----------------
CLASS_NAME = "player"           # change per session: player | ball | goal | background
DATASET_ROOT = "field_trip_1"   # a separate root folder for field data
 
BLUR_THRESHOLD = 50.0           # classroom default was 80. Lower outdoors (speed, wind, distance)
CAPTURE_INTERVAL = 0.0          # 0 = SPACE only. Switch to 0.8 for passive capture
MIN_DIFF = 800                  # higher than the classroom 500: the pitch background is static
CAMERA_INDEX = 0                # test before leaving. Use 1 or 2 if it fails
SAVE_QUALITY = 92               # keep high. Compress later if storage is short
MOTION_ONLY = False             # True: only try to save when something moves
 
TARGET_COUNTS = {
    "player": 80,               # majority class, easy to collect
    "ball": 80,                 # priority: a small object seen briefly, so it needs the same effort
    "goal": 40,                 # static object: capture first, then move on
    "background": 40,           # empty pitch sections as null examples
}
# Total target: 240 frames captured, about 150 after the filters (at about 60% accept rate)

Question: why is the ball target as high as the player target even though a ball is a tiny part of every image?

Team roles for Field Trip 1

Each team of 3 or 4 assigns these roles before arriving at the field. Roles rotate every 30 minutes.

In a team of three, the Logger and the Spotter share a role.

Your Field Collection Brief

Share as a Google Doc, one per team. Maximum 1 page when printed. It must be approved before Field Trip 1 (lesson 28).

Section 1: Class capture targets

Section 2: Camera positions

Section 3: capture.py configuration

Section 4: Team role assignment

Section 5: Our 5 non-negotiable quality rules

Write 5 rules your team will enforce. Each must trace back to a lesson from lesson 10 to lesson 14. Example: "We will capture at least 80 ball frames, because the ball is small and seen briefly, so it needs the same effort as the players."

Section 6: Consent and privacy (from lesson 14)

Section 7: Contingency plans

What will you do if each of these happens at the field?

What you'll need

• Your capture.py from lesson 11
• Your team (3 or 4 people)
• The Field Collection Brief template below

Build it

A planning calculator

You check whether their targets fit in the time they have. It's a small script that turns the plan into minutes.

def plan(targets, accept_rate=0.6, tries_per_minute=6, minutes_available=90, switch_minutes=2):
    """
    targets: frames you want KEPT per class, e.g. {"player": 80, "ball": 80}
    accept_rate: share of tries that pass the blur and duplicate filters (0 to 1)
    tries_per_minute: how many times a minute you can realistically try a capture by hand (SPACE)
    """
    kept_per_minute = accept_rate * tries_per_minute
    rows = []
    total_minutes = 0.0
    for cls, n in targets.items():
        minutes = n / kept_per_minute + switch_minutes
        total_minutes += minutes
        rows.append((cls, n, round(n / accept_rate), round(minutes, 1)))
 
    print(f"{'class':<12}{'keep':>6}{'try':>6}{'minutes':>9}")
    for cls, keep, tries, minutes in rows:
        print(f"{cls:<12}{keep:>6}{tries:>6}{minutes:>9}")
    print(f"Total: {round(total_minutes, 1)} minutes of {minutes_available} available")
    if total_minutes > minutes_available:
        print("Too much for the time. Lower a target, or find a way to raise the accept rate.")
    else:
        spare = round(minutes_available - total_minutes, 1)
        print(f"Fits, with {spare} minutes to spare for moving between positions and surprises.")
    return total_minutes
 
 
plan({"player": 80, "ball": 80, "goal": 40, "background": 40})
print()
plan({"player": 80, "ball": 80, "goal": 40, "background": 40}, accept_rate=0.3)

Question: what happens to the plan if the accept rate falls from 60% to 30%? Which section of the brief tells you what to do?

If it goes wrong

Try it

• Work out the shutter time needed to keep a ball travelling at 50 km/h within 5 pixels of blur at 10 m distance
• Draw a pitch diagram with the camera positions, field of view, and distance to the near and far players
• Write a step-by-step "capture protocol" for the Class Manager

Conclusion

A field trip without a plan brings back unusable footage. Your Field Collection Brief sets camera positions, per-class targets, capture settings, team roles and backup plans, and every rule in it traces back to a lesson from Unit 3.

What next?

Next: How neural networks learn. Unit 4 starts: how neural networks actually learn.