TASKS:
1.1
- Object Detection( a) Cars and b) Bikes in the Video)
1.2
- Lane Detection
Bonus
- Object Tracking (Bonus ) and Counting( extra from my side)
I have used the YOLOv5 algorithm for object detection for the given input video. Taken help from the following sources: guide: https://docs.ultralytics.com/quick-start/ reference video : https://github.com/ultralytics/yolov5
RESULTS: https://drive.google.com/file/d/1zimiVEmHrjT0--5r-wJQ38Sk2NS7VSUR/view?usp=sharing
NOTE: I understand the fact that YOLOv4 could have been the better choice wrt accuracy ,but due to some compatibility(most probably) issues , I wasn't able to complete that model using YOLOv4 but instead used YOLOv5 https://arxiv.org/pdf/2004.10934v1.pdf
In this part I have used :
- Camera Calibration -remove the curved line related errors present in video
- Thresholding - converting image to binary using some threshold value (To differenciate the lanes from surrounding)
- Perspective transformation ( considering the lanes to be going farther and thereby deciding the lines trapezoid like thing is actually a rectangle)
- LaneLines ( identifying the lane lines)
- etc
RESULTS: https://drive.google.com/file/d/1NdFZXxhFd9rXld62VRMdMql_XiXXXEMR/view?usp=sharing
references https://github.com/Dt-Pham/Advanced-Lane-Lines.git https://www.youtube.com/watch?v=iRTuCYx6quQjec
readings: https://viso.ai/deep-learning/object-tracking/
In that process I actually Solved a bug in the code, from the reference repository I was refering to ( basically the module it was using was outdated) https://stackoverflow.com/questions/62390517/no-module-named-sklearn-utils-linear-assignment
Attempted in the following manner : -used open cv based methods for object tracking
Added along with Tracker( My attempt for counting vehicles seems a lot inaccurate , I am trying to improve on it)
https://github.com/guptavasu1213/Yolo-Vehicle-Counter/blob/master/README.md https://github.com/bamwani/car-counting-and-speed-estimation-yolo-sort-python/blob/master/README.md