CV and Biomechanics

IN PYTHON:

# Step 1: Mount Google Drive

from google.colab import drive

drive.mount('/content/drive')

# Step 2: Define the path to the video file (adjust path based on where it's located in your Google Drive)

video_path = "/content/drive/MyDrive/GC.mp4"

# Now, you can use `video_path` to load the video in your previous code

import cv2

import pandas as pd

import mediapipe as mp

from google.colab.patches import cv2_imshow

# Initialize empty list to store joint data

joint_data = []

# Load the video

cap = cv2.VideoCapture(video_path)

# Check if video opened successfully

if not cap.isOpened():

    print(f"Error: Could not open video at {video_path}")

    exit()

# Initialize MediaPipe Pose

mp_pose = mp.solutions.pose

pose = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.5, min_tracking_confidence=0.5)

frame_count = 0

while cap.isOpened():

    ret, frame = cap.read()

    if not ret:

        break

    # Process only every 4th frame

    if frame_count % 4 == 0:

        # Convert the BGR frame to RGB

        rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)

        results = pose.process(rgb_frame)

        # If any landmarks are found

        if results.pose_landmarks:

            # Draw landmarks on the frame

            mp.solutions.drawing_utils.draw_landmarks(frame, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)

            # Extract joint positions (shoulder, elbow, wrist)

            shoulder = (int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].x * frame.shape[1]),

                        int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].y * frame.shape[0]))

            elbow = (int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].x * frame.shape[1]),

                      int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].y * frame.shape[0]))

            wrist = (int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].x * frame.shape[1]),

                      int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].y * frame.shape[0]))

            # Draw the points on the frame

            cv2.circle(frame, shoulder, 5, (0, 255, 0), -1)  # Green for shoulder

            cv2.circle(frame, elbow, 5, (255, 0, 0), -1)     # Blue for elbow

            cv2.circle(frame, wrist, 5, (0, 0, 255), -1)      # Red for wrist

            # Save the joint data for this frame

            joint_data.append({

                'frame': frame_count,

                'shoulder_x': shoulder[0],

                'shoulder_y': shoulder[1],

                'elbow_x': elbow[0],

                'elbow_y': elbow[1],

                'wrist_x': wrist[0],

                'wrist_y': wrist[1]

            })

        # Display the frame (optional, comment this out if not needed)

        cv2_imshow(frame)

    frame_count += 1

# Release the video capture object

cap.release()

# Convert joint data to a DataFrame and save it

joint_data_df = pd.DataFrame(joint_data)

joint_data_df.to_csv('joint_data.csv', index=False)

print("Joint data saved to 'joint_data.csv'")

IN R:

gerritcole <- read.csv("C://Users/Drew Duffy/Downloads/joint_data_1.csv")

colnames(gerritcole)

library(ggplot2)

library(gganimate)

# Create the plot

p <- ggplot(gerritcole, aes(frame = frame)) +

  # Left Shoulder trajectory and points

  geom_point(aes(x = shoulder_l_x, y = shoulder_l_y, color = "Left Shoulder"), size = 3) +

  geom_path(aes(x = shoulder_l_x, y = shoulder_l_y, group = 1, color = "Left Shoulder"), size = 1) +

  # Right Shoulder trajectory and points

  geom_point(aes(x = shoulder_r_x, y = shoulder_r_y, color = "Right Shoulder"), size = 3) +

  geom_path(aes(x = shoulder_r_x, y = shoulder_r_y, group = 1, color = "Right Shoulder"), size = 1) +

  # Left Elbow trajectory and points

  geom_point(aes(x = elbow_l_x, y = elbow_l_y, color = "Left Elbow"), size = 3) +

  geom_path(aes(x = elbow_l_x, y = elbow_l_y, group = 1, color = "Left Elbow"), size = 1) +

  # Right Elbow trajectory and points

  geom_point(aes(x = elbow_r_x, y = elbow_r_y, color = "Right Elbow"), size = 3) +

  geom_path(aes(x = elbow_r_x, y = elbow_r_y, group = 1, color = "Right Elbow"), size = 1) +

  # Left Wrist trajectory and points

  geom_point(aes(x = wrist_l_x, y = wrist_l_y, color = "Left Wrist"), size = 3) +

  geom_path(aes(x = wrist_l_x, y = wrist_l_y, group = 1, color = "Left Wrist"), size = 1) +

  # Right Wrist trajectory and points

  geom_point(aes(x = wrist_r_x, y = wrist_r_y, color = "Right Wrist"), size = 3) +

  geom_path(aes(x = wrist_r_x, y = wrist_r_y, group = 1, color = "Right Wrist"), size = 1) +

  # Left Knee trajectory and points

  geom_point(aes(x = knee_l_x, y = knee_l_y, color = "Left Knee"), size = 3) +

  geom_path(aes(x = knee_l_x, y = knee_l_y, group = 1, color = "Left Knee"), size = 1) +

  # Right Knee trajectory and points

  geom_point(aes(x = knee_r_x, y = knee_r_y, color = "Right Knee"), size = 3) +

  geom_path(aes(x = knee_r_x, y = knee_r_y, group = 1, color = "Right Knee"), size = 1) +

  # Animation settings

  transition_reveal(frame) +  # Smoothly reveal each frame

  theme_minimal() +

  # Title and axis labels

  labs(title = "2D Trajectory of Shoulders, Elbows, Wrists, and Knees: Frame {frame}",

       x = "X Coordinate", y = "Y Coordinate", color = "Joint") +

  # Slowing down the animation by adjusting frame duration

  ease_aes('linear') +

  theme(plot.title = element_text(size = 15))

# Save or animate

animate(p, duration = 10, fps = 100, width = 800, height = 600)

# Save or display the animation

anim_save("joint_trajectory.gif")

IN PYTHON:

# Step 1: Mount Google Drive

from google.colab import drive

drive.mount('/content/drive')

# Step 2: Define the path to the video file

video_path = "/content/drive/MyDrive/vh.mp4"

import cv2

import pandas as pd

import mediapipe as mp

from google.colab.patches import cv2_imshow

import time

# Initialize empty list to store joint data

joint_data = []

# Load the video

cap = cv2.VideoCapture(video_path)

# Check if video opened successfully

if not cap.isOpened():

    print(f"Error: Could not open video at {video_path}")

    exit()

# Initialize MediaPipe Pose

mp_pose = mp.solutions.pose

pose = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.5, min_tracking_confidence=0.5)

frame_count = 0

fps = cap.get(cv2.CAP_PROP_FPS)

delay = int(1000 / fps)  # Delay in milliseconds for cv2.imshow()

# Process the video frame by frame

while cap.isOpened():

    ret, frame = cap.read()

    if not ret:

        break

    # Process every 4th frame (this will also slow it down)

    if frame_count % 7 == 0:

        # Convert the BGR frame to RGB

        rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)

        results = pose.process(rgb_frame)

        # If any landmarks are found

        if results.pose_landmarks:

            # Draw landmarks on the frame

            mp.solutions.drawing_utils.draw_landmarks(frame, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)

            # Extract joint positions (shoulders, elbows, wrists, hips, knees)

            joints = {

                'shoulder_l_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].x * frame.shape[1]),

                'shoulder_l_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_SHOULDER].y * frame.shape[0]),

                'shoulder_r_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].x * frame.shape[1]),

                'shoulder_r_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_SHOULDER].y * frame.shape[0]),

                'elbow_l_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].x * frame.shape[1]),

                'elbow_l_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_ELBOW].y * frame.shape[0]),

                'elbow_r_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].x * frame.shape[1]),

                'elbow_r_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_ELBOW].y * frame.shape[0]),

                'wrist_l_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].x * frame.shape[1]),

                'wrist_l_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_WRIST].y * frame.shape[0]),

                'wrist_r_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].x * frame.shape[1]),

                'wrist_r_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_WRIST].y * frame.shape[0]),

                'hip_l_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].x * frame.shape[1]),

                'hip_l_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_HIP].y * frame.shape[0]),

                'hip_r_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].x * frame.shape[1]),

                'hip_r_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_HIP].y * frame.shape[0]),

                'knee_l_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_KNEE].x * frame.shape[1]),

                'knee_l_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.LEFT_KNEE].y * frame.shape[0]),

                'knee_r_x': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_KNEE].x * frame.shape[1]),

                'knee_r_y': int(results.pose_landmarks.landmark[mp_pose.PoseLandmark.RIGHT_KNEE].y * frame.shape[0]),

            }

            # Save the joint data for this frame

            joint_data.append({'frame': frame_count, **joints})

        # Display the frame (optional, comment this out if not needed)

        cv2_imshow(frame)

    frame_count += 1

    # Slowing down the processing

    time.sleep(0.1)  # Adjust this to slow down the playback more if needed

# Release the video capture object

cap.release()

# Convert joint data to a DataFrame and save it

joint_data_df = pd.DataFrame(joint_data)

joint_data_df.to_csv('/content/drive/MyDrive/joint_data.csv', index=False)

print("Joint data saved to 'joint_data.csv'")

IN R:

vance <- read.csv("C://Users/Drew Duffy/Downloads/vh2.csv")

library(dplyr)

library(tidyr)

library(ggplot2)

swing_indices <- list(

  swing1 = c(1, 6, 7, 8, 9, 10),

  swing2 = c(14, 16, 17, 18, 29, 20),

  swing3 = c(25, 26, 27, 28, 29)

)

calculate_swing_metrics <- function(data, indices) {

  swing_data <- data %>% slice(indices)

  metrics <- swing_data %>%

    summarise(

      wrist_speed = sqrt(mean((wrist_r_x - wrist_l_x)^2 + (wrist_r_y - wrist_l_y)^2)),

      elbow_angle = sqrt(mean((elbow_r_x - elbow_l_x)^2 + (elbow_r_y - elbow_l_y)^2)),

      hip_angle = sqrt(mean((hip_r_x - hip_l_x)^2 + (hip_r_y - hip_l_y)^2)),

      knee_angle = sqrt(mean((knee_r_x - knee_l_x)^2 + (knee_r_y - knee_l_y)^2))

    )

  return(metrics)

}

swing_metrics <- lapply(swing_indices, function(indices) calculate_swing_metrics(vance, indices))

swing_metrics_df <- do.call(rbind, swing_metrics) %>%

  mutate(swing = row_number())

print(swing_metrics_df)

swing_metrics_long <- swing_metrics_df %>%

  pivot_longer(cols = -swing, names_to = "metric", values_to = "value")

ggplot(swing_metrics_long, aes(x = factor(swing), y = value, fill = metric)) +

  geom_bar(stat = "identity", position = position_dodge()) +

  labs(title = "Swing Metrics for Vance's 3 Swings", x = "Swing", y = "Value") +

  theme_minimal() +

  theme(legend.position = "bottom") +

  scale_fill_brewer(palette = "Set1")

swing_indices <- list(

  swing1 = c(1, 6, 7, 8, 9, 10, 11),

  swing2 = c(14, 16, 17, 18, 19, 20),  # Fixed index from 29 to 19

  swing3 = c(25, 26, 27, 28, 29, 30)

)

calculate_velocity <- function(data, part_x, part_y) {

  velocities <- data %>%

    mutate(

      velocity_x = c(0, diff(get(part_x))),  # Calculate difference in x-coordinates

      velocity_y = c(0, diff(get(part_y)))   # Calculate difference in y-coordinates

    ) %>%

    rowwise() %>%

    mutate(velocity = sqrt(velocity_x^2 + velocity_y^2))  # Calculate resultant velocity

  return(velocities$velocity)

}

swing_velocities <- list()

for (swing in names(swing_indices)) {

  indices <- swing_indices[[swing]]

  swing_data <- vance %>% slice(indices)  # Get data for the current swing

  swing_velocities[[swing]] <- swing_data %>%

    mutate(

      wrist_velocity = calculate_velocity(swing_data, "wrist_r_x", "wrist_r_y"),

      elbow_velocity = calculate_velocity(swing_data, "elbow_r_x", "elbow_r_y"),

      hip_velocity = calculate_velocity(swing_data, "hip_r_x", "hip_r_y"),

      knee_velocity = calculate_velocity(swing_data, "knee_r_x", "knee_r_y")

    )

}

swing_velocity_combined <- do.call(rbind, lapply(swing_velocities, function(x) x %>% select(frame, wrist_velocity, elbow_velocity, hip_velocity, knee_velocity)))

velocity_data_long <- swing_velocity_combined %>%

  pivot_longer(cols = -frame, names_to = "body_part", values_to = "velocity")

ggplot(velocity_data_long, aes(x = frame, y = velocity, color = body_part)) +

  geom_line(size = 1) +

  labs(title = "Velocities of Body Parts for Vance's Swings", x = "Frame", y = "Velocity (units/frame)") +

  theme_minimal() +

  theme(legend.position = "bottom") +

  facet_wrap(~body_part) +

  scale_color_brewer(palette = "Set1")