Field Trip 1: Capturing Football Data on a Real Pitch

Introduction

Today the classroom becomes a football field. With your Field Collection Brief from lesson 15, your team sets up a laptop and webcam beside a real pitch and collects a dataset of players, the ball, goals and the empty pitch. The skill isn't just pressing record: it's noticing what's going wrong and fixing it while you're still there.

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

• Carry out a data collection plan they wrote themselves, in real conditions
• Work in the roles from the field brief (Operator, Class Manager, Spotter, Logger and Reviewer)
• Change camera positions to create several sessions per class
• Use field_check.py to find what is missing while they are still at the field
• Follow the consent and privacy rules they wrote (lesson 14)
• Adapt when something goes wrong, and record what happened
• Back up the data and write a short trip report

Working principle

The plan is the plan, until it isn't

The approved brief says what to do. At the field the sun, the wind and the players will change things. The skill today is to notice (the on-screen blur number, the accept rate, the log) and adjust (a different angle, a lower threshold, a break in the shade) while keeping the rules you wrote.

Sessions make the dataset useful

Each time you change the camera position, press N to start a new session. In lesson 13 we will keep whole sessions together in one split, so we need several sessions per class. A target of 80 ball frames from a single position is much less useful than 80 from four positions.

Check while you can still fix it

The most expensive discovery is "we have no goal photos" back in the classroom. So at the halfway point, and again at the end, the Logger runs field_check.py and the team reacts.

Consent, again

The consent rules from lesson 14 apply at the field. Signed forms for every child in frame. The camera points at the pitch and not the stands. If anyone objects, stop, delete their frames and move. Your teacher or team leader carries the list.

What you'll need

• Your approved Field Collection Brief (lesson 15)
• A charged laptop, a webcam on a stand, and capture.py set to your field settings
• field_check.py (below)
• Water, a hat and sunscreen

Preparation

Team checklist (the day before)

Plan for the day

A suggested 2-hour schedule. Adjust to your pitch and light.

Roles (from lesson 15)

Rotate roles at Checkpoint 1 so everyone has a turn at the laptop.

Field tools

field_check.py: are we on track?

Put this in the folder that contains your DATASET_ROOT. Edit TARGETS to match your brief. Run it with python field_check.py field_trip_1, using the root folder name from capture.py. It reads the CSV log that capture.py writes (lesson 11), so it needs no camera.

import csv
import os
import sys
from collections import Counter, defaultdict
 
ROOT = sys.argv[1] if len(sys.argv) > 1 else "field_trip_1"        # DATASET_ROOT in capture.py
TARGETS = {"player": 80, "ball": 80, "goal": 40, "background": 40}   # copy these from your approved brief
MIN_ACCEPT_RATE = 0.40
MIN_SESSIONS_PER_CLASS = 3
 
log_path = os.path.join(ROOT, "capture_log.csv")
if not os.path.exists(log_path):
    raise SystemExit(f"No log at {log_path}. Is DATASET_ROOT right?")
 
with open(log_path) as f:
    rows = list(csv.DictReader(f))
 
tries = len(rows)
kept = [r for r in rows if r["accepted"] == "True"]
print(f"Attempts: {tries}   Kept: {len(kept)}   Accept rate: {len(kept) / max(tries, 1):.0%}")
 
reasons = Counter(r["reason"] for r in rows if r["accepted"] != "True")
if reasons:
    print("Rejected because:", ", ".join(f"{k} {v}" for k, v in reasons.most_common()))
 
sessions = defaultdict(set)
per_class = Counter()
for r in kept:
    per_class[r["class"]] += 1
    sessions[r["class"]].add(r["session"])
 
print(f"\n{'class':<12}{'kept':>6}{'target':>8}{'sessions':>10}   status")
warnings = []
for cls, target in TARGETS.items():
    n, s = per_class[cls], len(sessions[cls])
    status = "OK"
    if n < target:
        status = f"NEED {target - n} MORE"
        warnings.append(f"{cls}: {n}/{target} frames")
    if s < MIN_SESSIONS_PER_CLASS:
        status += f"  (only {s} session{'s' if s != 1 else ''}: change position!)"
        warnings.append(f"{cls}: only {s} session(s), the lesson 13 split needs variety")
    print(f"{cls:<12}{n:>6}{target:>8}{s:>10}   {status}")
 
# Do the files on disk match the log?
on_disk = Counter()
for cls in TARGETS:
    folder = os.path.join(ROOT, cls)
    if os.path.isdir(folder):
        on_disk[cls] = len([x for x in os.listdir(folder) if x.endswith(".jpg")])
mismatch = [c for c in TARGETS if on_disk[c] != per_class[c]]
if mismatch:
    warnings.append("files and log disagree for: " + ", ".join(mismatch))
 
if len(kept) / max(tries, 1) < MIN_ACCEPT_RATE:
    warnings.append(f"accept rate below {MIN_ACCEPT_RATE:.0%}: check blur number, camera shake, MIN_DIFF")
 
print("\nTOTAL KEPT:", len(kept), "(goal: 200 or more across the 4 classes)")
print("ALL CLEAR" if not warnings else "FIX BEFORE LEAVING THE FIELD:")
for w in warnings:
    print("  -", w)

It tells you, for each class: how many frames you kept, your target, and in how many sessions. It flags classes with too few frames or too few positions, and any mismatch between the files and the log.

Quality rules of thumb at the field

After the field

Data handling (same day)

1. Copy the DATASET_ROOT folder to a second location (a USB drive, or the school drive). Do this before anything else.
2. Run field_check.py one last time and save its output.
3. Do not post raw images in group chats or public places (lesson 14).
4. Upload to the labelling tool only if your charter allows it, and only with consent in place.
5. Write the deletion date from the consent form on the folder.

Trip report: template

Each team submits one page in the the next lesson.

If it goes wrong

Try it

• Plot a histogram of blur scores per position from the log and bring it to the report
• Propose a better position by sketching the pitch with the camera's field of view
• Time how long each class took to reach its target

Conclusion

In the field, the plan meets reality. Change position and start a new session often, watch the blur score and accept rate, check your class counts halfway with field_check.py, keep to the consent rules, and back the dataset up before you leave.

What next?

Next: Clean the field data & retrain. Back in the classroom, you'll clean the catch and retrain on real field data.