Differences

This shows you the differences between two versions of the page.

--- amc:ss2025:group-a:start [2025/07/29 11:27] – [Pictures of the prototype] 35120_students.hsrw
+++ amc:ss2025:group-a:start [2025/07/29 15:11] (current) – [Data analysis] 35120_students.hsrw
@@ Line 1: / Line 1: @@
+====== Gießwagen - Plant Detection ======
+By Dylan Elian Huete Arbizu (35120)
 ==== Introduction ====
@@ Line 31: / Line 34: @@
   * Wi-Fi Network: For ESP32 to connect and transmit data.
-===== Pin Assignments =====
+==== Pin Assignments ====
 ^ Function                          ^ ESP32 GPIO  ^ Notes                          ^
@@ Line 795: / Line 798: @@
 ==== Data analysis ====
+**Data Cleaning:**
+Selection of a segment of interest, discarding outliers or unreliable data.
+Due some hardware limitations, the reading of the marks was not reliable enough, therefore the data had to be conditioned manually.
+The expected data is 12 marks readings, but due to problems in the data acquisition there were just 8 marks usable, and there were also some duplicated readings, this was determined manually based on the time and pattern expected.
+**Path Segmentation:**
+getPath(df) constructs segments ("paths") marked by zero in the zone_1 column, grouping each set of four zeros as a new path.
+**Speed Calculation:**
+mean_speed(data) estimates the average speed between time marks by measuring intervals between zeros.
+  * After interpolating sensor data along the location axis (temporal/spatial sequence), each 8×8 frame is upscaled to 64×64 pixels using cubic spline interpolation (zoom with order=3).
+  * This spatial interpolation significantly enhances intra-frame resolution.
+  * The aggregation step then merges these larger frames horizontally with value averaging over overlapping columns, preserving continuity.
+  * The plot displays a much higher-resolution heatmap representing the sensor data over the scanned path.
+<code python>
+import pandas as pd
+import matplotlib.pyplot as plt
+import numpy as np
+from scipy.ndimage import zoom
+from scipy.interpolate import interp1d
+fname1 = "vl53l8cx_data_third_test.csv"
+df = pd.read_csv(fname1)
+#Since the marks acquisition is not reliable enough, the data has to be treated manually to discard unuseful data
+dataFrame = df.loc[1419:2618].drop(2360)
+def getPath(df):
+    idx = list(df.loc[df["zone_1"]==0].index)
+    path_list = []
+    for i in range(len(idx)):
+        if (i+1)%4 == 0 and i != 0:
+            path_list.append(df.loc[idx[i-3]:idx[i]])
+            #print(f"{i} : [{idx[i-3]}:{idx[i]}]")
+    return path_list
+def mean_speed(data):
+    idx = list(data.loc[data["zone_1"]==0].index)
+    t1 = data.loc[idx[1]]["timestamp_us"] - data.loc[idx[0]]["timestamp_us"]
+    t2 = data.loc[idx[2]]["timestamp_us"] - data.loc[idx[1]]["timestamp_us"]
+    t3 = data.loc[idx[3]]["timestamp_us"] - data.loc[idx[2]]["timestamp_us"]
+    spd1 = 200/t1
+    spd2 = 1000/t2
+    spd3 = 1000/t3
+    return (spd1+spd2+spd3)/3
+paths = getPath(dataFrame)
+# 1. Calculate continuous locations and concatenate all paths as before:
+for i in range(len(paths)):
+    loc = (paths[i]["timestamp_us"] - paths[i].iloc[0]["timestamp_us"]) * mean_speed(paths[i])
+    paths[i] = pd.concat([paths[i], loc.to_frame('location')], axis=1)
+    paths[i] = paths[i].drop(list(paths[i].loc[paths[i]["zone_1"] == 0].index))
+path_t = pd.concat(paths)
+path_t = path_t.sort_values("location")
+locations = path_t['location'].values
+data_values = path_t.iloc[:, 1:-1].values  # Adjust indices if your columns differ
+# 2. Interpolate sensor columns independently on a uniform location grid:
+min_loc, max_loc = np.min(locations), np.max(locations)
+num_interp_points = int(np.ceil(max_loc - min_loc)) + 1
+interp_locations = np.linspace(min_loc, max_loc, num_interp_points)
+interp_data = np.zeros((num_interp_points, data_values.shape[1]))
+for col in range(data_values.shape[1]):
+    interp_func = interp1d(locations, data_values[:, col], kind='linear', fill_value='extrapolate')
+    interp_data[:, col] = interp_func(interp_locations)
+# 3. Reshape each row into 8x8 frames:
+num_frames = interp_data.shape[0]
+frame_height, frame_width = 8, 8
+frames_8x8 = [interp_data[i].reshape(frame_height, frame_width) for i in range(num_frames)]
+# 4. Interpolate each 8x8 frame to 64x64 using scipy.ndimage.zoom:
+zoom_factor = 64 / 8  # 8x to 64x scaling
+frames_64x64 = [zoom(frame, zoom_factor, order=3) for frame in frames_8x8]  # cubic spline interpolation (order=3)
+# 5. Aggregate frames horizontally with averaging over overlaps (same as before):
+max_offset = num_frames - 1
+final_width = max_offset + 64  # width after scaling frames to 64 wide
+final_frame_64 = np.zeros((64, final_width))
+count_64 = np.zeros((64, final_width))
+for i, frame in enumerate(frames_64x64):
+    offset = i
+    final_frame_64[:, offset:offset+64] += frame
+    count_64[:, offset:offset+64] += 1
+aggregated_64 = np.divide(final_frame_64, count_64, out=np.zeros_like(final_frame_64), where=count_64 != 0)
+# 6. Plot the aggregated 64x wide frame:
+plt.figure(figsize=(final_width / 16, 8))  # Adjust size for clarity
+plt.imshow(aggregated_64, cmap='viridis', aspect='auto')
+plt.colorbar(label='Value')
+plt.title("Aggregated Large Frame with 8x8 to 64x64 Spatial Interpolation")
+plt.xlabel('Columns (scaled)')
+plt.ylabel('Rows (scaled)')
+plt.show()
+</code>
+{{ :amc:ss2025:group-a:screenshot_2025-07-29_142824.png?800 |Map of the path of the Gießwagen}}
 ===== Discussion =====