From 8ac6e71d481ef881ec0fcdfcec5c3db4fcd553af Mon Sep 17 00:00:00 2001
From: Mike Lang <mikelang3000@gmail.com>
Date: Fri, 17 Nov 2023 00:42:19 +1100
Subject: [PATCH] bus_analyzer: Normalize brightness across all digits, not one
 by one

---
 bus_analyzer/bus_analyzer/extract.py | 18 ++++++++++--------
 bus_analyzer/bus_analyzer/main.py    | 25 ++++++++++++++++---------
 2 files changed, 26 insertions(+), 17 deletions(-)

diff --git a/bus_analyzer/bus_analyzer/extract.py b/bus_analyzer/bus_analyzer/extract.py
index e221791..d060032 100644
--- a/bus_analyzer/bus_analyzer/extract.py
+++ b/bus_analyzer/bus_analyzer/extract.py
@@ -99,32 +99,34 @@ def extract_digits(image, type):
 		for dx in range(DIGIT_WIDTH)
 	]
 	main_digits = get_brightest_region(image, digit_xs, DIGIT_HEIGHT)
+	main_digits = normalize(main_digits)
 
 	digits = []
 	for i, x in enumerate(main_digit_coords):
 		digit = main_digits.crop((x, 0, x + DIGIT_WIDTH, main_digits.height))
-		digits.append(normalize_digit(digit))
+		digits.append(digit)
 
 	if type == "odo":
 		x = DIGIT_X_COORDS["odo"][-1]
 		last_digit = get_brightest_region(image, range(x, x + DIGIT_WIDTH), LAST_DIGIT_HEIGHT)
 		last_digit = last_digit.crop((x, 0, x + DIGIT_WIDTH, last_digit.height))
-		digits.append(normalize_digit(last_digit))
+		last_digit = normalize(last_digit)
+		digits.append(last_digit)
 
 	return digits
 
 
-def normalize_digit(digit):
+def normalize(image):
 	# Expand the range of the image so that the darkest pixel becomes black
 	# and the lightest becomes white
-	_min, _max = digit.getextrema()
+	_min, _max = image.getextrema()
 	_range = _max - _min
 	if _range == 0:
-		digit = digit.point(lambda v: 128)
+		image = image.point(lambda v: 128)
 	else:
-		digit = digit.point(lambda v: 255 * (v - _min) / _range)
+		image = image.point(lambda v: 255 * (v - _min) / _range)
 	
-	return digit
+	return image
 
 
 def recognize_digit(prototypes, image, blank_is_zero=False):
@@ -181,7 +183,7 @@ def read_digit(digit, prototypes_path="./prototypes", verbose=False):
 	guess, score, all_scores = recognize_digit(prototypes["odo-digits"], digit)
 	print("Digit = {} with score {}".format(guess, score))
 	if verbose:
-		all_scores.sort(key=lambda x: x[1])
+		all_scores.sort(key=lambda x: -1 if x[1] is None else x[1])
 		for s, n in all_scores:
 			print("{}: {}".format(n, s))
 
diff --git a/bus_analyzer/bus_analyzer/main.py b/bus_analyzer/bus_analyzer/main.py
index 1124bc2..c83a20a 100644
--- a/bus_analyzer/bus_analyzer/main.py
+++ b/bus_analyzer/bus_analyzer/main.py
@@ -48,7 +48,7 @@ def compare_segments(dbconnect, base_dir='.', prototypes_path="./prototypes", si
 		where = ["true"]
 	where = " AND ".join(where)
 	result = database.query(conn, f"""
-		SELECT odometer, segment
+		SELECT odometer, clock, timeofday, segment
 		FROM bus_data
 		WHERE segment IS NOT NULL
 			AND {where}
@@ -57,7 +57,7 @@ def compare_segments(dbconnect, base_dir='.', prototypes_path="./prototypes", si
 	# To get a wider range of tests, pick at random from all unique odo readings
 	available = {}
 	for row in result.fetchall():
-		available.setdefault(row.odometer, []).append(row.segment)
+		available.setdefault(row.odometer, []).append((row.segment, row.clock, row.timeofday))
 
 	selected = []
 	while available and len(selected) < num:
@@ -72,18 +72,25 @@ def compare_segments(dbconnect, base_dir='.', prototypes_path="./prototypes", si
 			del available[odometer]
 
 	results = []
-	for old_odometer, segment in selected:
+	for old_odometer, (segment, old_clock, old_tod) in selected:
 		path = os.path.join(base_dir, segment)
 		segment_info = parse_segment_path(path)
 		odometer, clock, tod = extract_segment(prototypes, segment_info)
-		results.append((segment, old_odometer, odometer))
+		results.append((segment, {
+			"odo": (old_odometer, odometer),
+			"clock": (old_clock, clock),
+			"tod": (old_tod, tod),
+		}))
 
 	matching = 0
-	for segment, old_odometer, odometer in sorted(results, key=lambda t: t[0]):
-		match = old_odometer == odometer
-		if verbose or not match:
-			print(f"{segment}: {old_odometer} | {odometer}")
-		if match:
+	for segment, data in sorted(results, key=lambda t: t[0]):
+		matched = True
+		for k, (old, new) in data.items():
+			match = old == new
+			if verbose or not match:
+				print(f"{segment} {k}: {old} | {new}")
+			matched &= match
+		if matched:
 			matching += 1
 
 	print("{}/{} matched".format(matching, len(selected)))