Added demo2

README.md

@@ -1,14 +1,16 @@
 # BasketTracking
 
-We aim at developing a system capable of 
-action tracking and understanding in basketball games using 
-different computer vision approaches and ideas alongside 
-deep learning models such as Detectron2. 
+We aim at developing a system capable of action tracking and understanding in basketball games using different computer
+vision approaches and ideas alongside deep learning models such as Detectron2.
 
+### Demo
+
+<p align='center'>
+ <img src="demo.gif" width="90%"/>
+</p> 
 
 
-### Demo
 <p align='center'>
- <img src="demo.gif" />
+ <img src="demo2.gif" width="90%"/>
 </p>
 

ball_detect_track.py

@@ -84,10 +84,10 @@ def ball_tracker(self, M, M1, frame, map_2d, map_2d_text, timestamp):
  ball_center = np.array([int(bbox[0] + bbox[2] / 2), int(bbox[1] + bbox[3] / 2), 1])
  clean_frame = frame.copy()
 
- bbox_iou = (ball_center[1]-IOU_BALL_PADDING,
- ball_center[0]- IOU_BALL_PADDING,
- ball_center[1]+IOU_BALL_PADDING,
- ball_center[0]+IOU_BALL_PADDING)
+ bbox_iou = (ball_center[1] - IOU_BALL_PADDING,
+ ball_center[0] - IOU_BALL_PADDING,
+ ball_center[1] + IOU_BALL_PADDING,
+ ball_center[0] + IOU_BALL_PADDING)
  scores = []
 
  for p in self.players:

feet_detect.py

@@ -70,7 +70,6 @@ def bb_intersection_over_union(boxA, boxB):
  # return the intersection over union value
  return iou
 
-
  def get_players_pos(self, M, M1, frame, timestamp, map_2d):
  warped_kpts = []
  outputs_seg = self.predictor_seg(frame)
@@ -86,7 +85,8 @@ def get_players_pos(self, M, M1, frame, timestamp, map_2d):
 
  for i, entry in enumerate(indices): # picking only class 0 (people)
  if entry == 0:
- ppl.append(np.array(cv2.erode(np.array(predicted_masks[i], dtype=np.uint8), kernel, iterations=4), dtype=bool))
+ ppl.append(
+ np.array(cv2.erode(np.array(predicted_masks[i], dtype=np.uint8), kernel, iterations=4), dtype=bool))
 
  indexes_ppl = np.array(
  [np.array(np.where(p == True)).T for p in ppl])

main.py

@@ -88,7 +88,7 @@ def get_frames(video_path, central_frame, mod):
  resized = cv2.resize(rectified, (map.shape[1], map.shape[0]))
  map = cv2.resize(map, (rectified.shape[1], rectified.shape[0]))
 
- video = cv2.VideoCapture("resources/Short4Mosaicing.mp4")
+ video = cv2.VideoCapture("resources/Short4Mosaicing2.mp4")
 
  players = []
  for i in range(1, 6):

video_handler.py

@@ -23,7 +23,7 @@ def __init__(self, pano, video, ball_detector, feet_detector, map_2d):
  self.map_2d = map_2d
 
  def run_detectors(self):
- writer = skvideo.io.FFmpegWriter("demo.mp4")
+ writer = skvideo.io.FFmpegWriter("demo2.mp4")
  time_index = 0
  while self.video.isOpened():
  ok, frame = self.video.read()
@@ -39,7 +39,8 @@ def run_detectors(self):
  M = self.get_homography(frame, self.des1, self.kp1)
  frame, self.map_2d, map_2d_text = self.feet_detector.get_players_pos(M, self.M1, frame, time_index,
  self.map_2d)
- frame, ball_map_2d = self.ball_detector.ball_tracker(M, self.M1, frame, self.map_2d.copy(), map_2d_text, time_index)
+ frame, ball_map_2d = self.ball_detector.ball_tracker(M, self.M1, frame, self.map_2d.copy(),
+ map_2d_text, time_index)
  vis = np.vstack((frame, cv2.resize(map_2d_text, (frame.shape[1], frame.shape[1] // 2))))
 
  cv2.imshow("Tracking", vis)