Spaces:
Sleeping
Sleeping
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,210 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import numpy as np
|
3 |
+
import torch
|
4 |
+
from ultralytics import YOLO
|
5 |
+
import gradio as gr
|
6 |
+
from scipy.interpolate import interp1d
|
7 |
+
import plotly.graph_objects as go
|
8 |
+
import uuid
|
9 |
+
import os
|
10 |
+
import tempfile
|
11 |
+
|
12 |
+
# Load YOLOv8 model and resolve class index
|
13 |
+
model = YOLO("best.pt")
|
14 |
+
model.to('cuda' if torch.cuda.is_available() else 'cpu')
|
15 |
+
|
16 |
+
# Dynamically resolve ball class index
|
17 |
+
ball_class_index = None
|
18 |
+
for k, v in model.names.items():
|
19 |
+
if v.lower() == "cricketball":
|
20 |
+
ball_class_index = k
|
21 |
+
break
|
22 |
+
if ball_class_index is None:
|
23 |
+
raise ValueError("Class 'cricketBall' not found in model.names")
|
24 |
+
|
25 |
+
# Constants
|
26 |
+
STUMPS_WIDTH = 0.2286
|
27 |
+
BALL_DIAMETER = 0.073
|
28 |
+
FRAME_RATE = 20
|
29 |
+
SLOW_MOTION_FACTOR = 3
|
30 |
+
CONF_THRESHOLD = 0.2
|
31 |
+
IMPACT_ZONE_Y = 0.85
|
32 |
+
IMPACT_DELTA_Y = 50
|
33 |
+
PITCH_LENGTH = 20.12
|
34 |
+
STUMPS_HEIGHT = 0.71
|
35 |
+
MAX_POSITION_JUMP = 30
|
36 |
+
|
37 |
+
def process_video(video_path):
|
38 |
+
if not os.path.exists(video_path):
|
39 |
+
return [], [], [], "Error: Video file not found"
|
40 |
+
cap = cv2.VideoCapture(video_path)
|
41 |
+
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
42 |
+
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
43 |
+
frames, ball_positions, detection_frames, debug_log = [], [], [], []
|
44 |
+
frame_count = 0
|
45 |
+
|
46 |
+
while cap.isOpened():
|
47 |
+
ret, frame = cap.read()
|
48 |
+
if not ret:
|
49 |
+
break
|
50 |
+
frame_count += 1
|
51 |
+
frames.append(frame.copy())
|
52 |
+
results = model.predict(frame, conf=CONF_THRESHOLD, imgsz=(frame_height, frame_width), iou=0.5, max_det=1)
|
53 |
+
detections = 0
|
54 |
+
for detection in results[0].boxes:
|
55 |
+
if int(detection.cls) == ball_class_index:
|
56 |
+
detections += 1
|
57 |
+
if detections == 1:
|
58 |
+
x1, y1, x2, y2 = detection.xyxy[0].cpu().numpy()
|
59 |
+
ball_positions.append([(x1 + x2) / 2, (y1 + y2) / 2])
|
60 |
+
detection_frames.append(frame_count - 1)
|
61 |
+
cv2.rectangle(frame, (int(x1), int(y1)), (int(x2), int(y2)), (0, 255, 0), 2)
|
62 |
+
frames[-1] = frame
|
63 |
+
debug_log.append(f"Frame {frame_count}: {detections} ball detections")
|
64 |
+
cap.release()
|
65 |
+
|
66 |
+
if not ball_positions:
|
67 |
+
debug_log.append("No balls detected in any frame")
|
68 |
+
else:
|
69 |
+
debug_log.append(f"Total ball detections: {len(ball_positions)}")
|
70 |
+
debug_log.append(f"Video resolution: {frame_width}x{frame_height}")
|
71 |
+
|
72 |
+
return frames, ball_positions, detection_frames, "\n".join(debug_log)
|
73 |
+
|
74 |
+
def find_bounce_point(ball_coords):
|
75 |
+
"""
|
76 |
+
Detect bounce point using y-derivative reversal with early-frame suppression.
|
77 |
+
Looks for where y increases then decreases (ball hits ground).
|
78 |
+
"""
|
79 |
+
y_coords = [p[1] for p in ball_coords]
|
80 |
+
min_index = None
|
81 |
+
|
82 |
+
for i in range(2, len(y_coords) - 2):
|
83 |
+
dy1 = y_coords[i] - y_coords[i - 1]
|
84 |
+
dy2 = y_coords[i + 1] - y_coords[i]
|
85 |
+
if dy1 > 0 and dy2 < 0:
|
86 |
+
if i > len(y_coords) * 0.2:
|
87 |
+
min_index = i
|
88 |
+
break
|
89 |
+
|
90 |
+
if min_index is not None:
|
91 |
+
return ball_coords[min_index]
|
92 |
+
|
93 |
+
return ball_coords[len(ball_coords)//2]
|
94 |
+
|
95 |
+
def lbw_decision(ball_positions, trajectory, frames, pitch_point, impact_point):
|
96 |
+
if not frames or not trajectory or len(ball_positions) < 2:
|
97 |
+
return "Not enough data", trajectory, pitch_point, impact_point
|
98 |
+
|
99 |
+
frame_height, frame_width = frames[0].shape[:2]
|
100 |
+
stumps_x = frame_width / 2
|
101 |
+
stumps_y = frame_height * 0.9
|
102 |
+
stumps_width_pixels = frame_width * (STUMPS_WIDTH / 3.0)
|
103 |
+
|
104 |
+
pitch_x, _ = pitch_point
|
105 |
+
impact_x, impact_y = impact_point
|
106 |
+
|
107 |
+
if pitch_x < stumps_x - stumps_width_pixels / 2 or pitch_x > stumps_x + stumps_width_pixels / 2:
|
108 |
+
return f"Not Out (Pitched outside line)", trajectory, pitch_point, impact_point
|
109 |
+
if impact_x < stumps_x - stumps_width_pixels / 2 or impact_x > stumps_x + stumps_width_pixels / 2:
|
110 |
+
return f"Not Out (Impact outside line)", trajectory, pitch_point, impact_point
|
111 |
+
for x, y in trajectory:
|
112 |
+
if abs(x - stumps_x) < stumps_width_pixels / 2 and abs(y - stumps_y) < frame_height * 0.1:
|
113 |
+
return f"Out (Ball projected to hit stumps)", trajectory, pitch_point, impact_point
|
114 |
+
return f"Not Out (Missing stumps)", trajectory, pitch_point, impact_point
|
115 |
+
|
116 |
+
def estimate_trajectory(ball_positions, detection_frames, frame_height, frame_width):
|
117 |
+
if len(ball_positions) < 2:
|
118 |
+
return None, None, None, "Error: Not enough ball detections"
|
119 |
+
|
120 |
+
filtered_positions = [ball_positions[0]]
|
121 |
+
filtered_frames = [detection_frames[0]]
|
122 |
+
for i in range(1, len(ball_positions)):
|
123 |
+
prev, curr = filtered_positions[-1], ball_positions[i]
|
124 |
+
if np.linalg.norm(np.array(curr) - np.array(prev)) <= MAX_POSITION_JUMP:
|
125 |
+
filtered_positions.append(curr)
|
126 |
+
filtered_frames.append(detection_frames[i])
|
127 |
+
|
128 |
+
if len(filtered_positions) < 2:
|
129 |
+
return None, None, None, "Error: Filtered detections too few"
|
130 |
+
|
131 |
+
x_vals = [p[0] for p in filtered_positions]
|
132 |
+
y_vals = [p[1] for p in filtered_positions]
|
133 |
+
times = np.array(filtered_frames) / FRAME_RATE
|
134 |
+
|
135 |
+
try:
|
136 |
+
fx = interp1d(times, x_vals, kind='cubic', fill_value="extrapolate")
|
137 |
+
fy = interp1d(times, y_vals, kind='cubic', fill_value="extrapolate")
|
138 |
+
except Exception as e:
|
139 |
+
return None, None, None, f"Interpolation error: {str(e)}"
|
140 |
+
|
141 |
+
total_frames = max(filtered_frames) - min(filtered_frames) + 1
|
142 |
+
t_full = np.linspace(times[0], times[-1], max(5, total_frames * SLOW_MOTION_FACTOR))
|
143 |
+
x_full = fx(t_full)
|
144 |
+
y_full = fy(t_full)
|
145 |
+
trajectory = list(zip(x_full, y_full))
|
146 |
+
|
147 |
+
pitch_point = find_bounce_point(filtered_positions)
|
148 |
+
impact_point = filtered_positions[-1]
|
149 |
+
|
150 |
+
return trajectory, pitch_point, impact_point, "Trajectory estimated successfully"
|
151 |
+
|
152 |
+
def generate_replay(frames, trajectory, pitch_point, impact_point, detection_frames):
|
153 |
+
if not frames or not trajectory:
|
154 |
+
return None
|
155 |
+
|
156 |
+
temp_file = os.path.join(tempfile.gettempdir(), f"drs_output_{uuid.uuid4()}.mp4")
|
157 |
+
height, width = frames[0].shape[:2]
|
158 |
+
out = cv2.VideoWriter(temp_file, cv2.VideoWriter_fourcc(*'mp4v'), FRAME_RATE / SLOW_MOTION_FACTOR, (width, height))
|
159 |
+
|
160 |
+
min_frame = min(detection_frames)
|
161 |
+
max_frame = max(detection_frames)
|
162 |
+
total_frames = max_frame - min_frame + 1
|
163 |
+
traj_per_frame = max(1, len(trajectory) // total_frames)
|
164 |
+
indices = [min(i * traj_per_frame, len(trajectory)-1) for i in range(total_frames)]
|
165 |
+
|
166 |
+
for i, frame in enumerate(frames):
|
167 |
+
idx = i - min_frame
|
168 |
+
if 0 <= idx < len(indices):
|
169 |
+
end_idx = indices[idx]
|
170 |
+
points = np.array(trajectory[:end_idx+1], dtype=np.int32).reshape((-1, 1, 2))
|
171 |
+
cv2.polylines(frame, [points], False, (255, 0, 0), 2)
|
172 |
+
if pitch_point and i == detection_frames[0]:
|
173 |
+
cv2.circle(frame, tuple(map(int, pitch_point)), 6, (0, 0, 255), -1)
|
174 |
+
if impact_point and i == detection_frames[-1]:
|
175 |
+
cv2.circle(frame, tuple(map(int, impact_point)), 6, (0, 255, 255), -1)
|
176 |
+
for _ in range(SLOW_MOTION_FACTOR):
|
177 |
+
out.write(frame)
|
178 |
+
out.release()
|
179 |
+
return temp_file
|
180 |
+
|
181 |
+
def drs_review(video):
|
182 |
+
frames, ball_positions, detection_frames, debug_log = process_video(video)
|
183 |
+
if not frames or not ball_positions:
|
184 |
+
return "No frames or detections found.", None
|
185 |
+
|
186 |
+
frame_height, frame_width = frames[0].shape[:2]
|
187 |
+
trajectory, pitch_point, impact_point, log = estimate_trajectory(ball_positions, detection_frames, frame_height, frame_width)
|
188 |
+
if not trajectory:
|
189 |
+
return f"{log}\n{debug_log}", None
|
190 |
+
|
191 |
+
decision, _, _, _ = lbw_decision(ball_positions, trajectory, frames, pitch_point, impact_point)
|
192 |
+
replay_path = generate_replay(frames, trajectory, pitch_point, impact_point, detection_frames)
|
193 |
+
|
194 |
+
result_log = f"DRS Decision: {decision}\n\n{log}\n\n{debug_log}"
|
195 |
+
return result_log, replay_path
|
196 |
+
|
197 |
+
# Gradio Interface
|
198 |
+
iface = gr.Interface(
|
199 |
+
fn=drs_review,
|
200 |
+
inputs=gr.Video(label="Upload Cricket Delivery Video"),
|
201 |
+
outputs=[
|
202 |
+
gr.Textbox(label="DRS Result and Debug Info"),
|
203 |
+
gr.Video(label="Replay with Trajectory & Decision")
|
204 |
+
],
|
205 |
+
title="GullyDRS - AI-Powered LBW Review",
|
206 |
+
description="Upload a cricket delivery video. The system will track the ball, estimate trajectory, and return a replay with an OUT/NOT OUT decision."
|
207 |
+
)
|
208 |
+
|
209 |
+
if __name__ == "__main__":
|
210 |
+
iface.launch()
|