Driver Assist System

Frameworks and Language

• Python 3.9
• OpenCV
• Yolov5
• PyTorch

Code Structure

• Install all the requirements using pip install -r requirements.txt
• lane_detection.py is the main file that acts as starting point and does all the processing. Run it using python3 lane-detecion.py
• FrameProcessing.py does all video and image frame related processing.
• YoloTransformation.py does the job of object detection.
• cleanup.sh cleans up the workspace.

To run the program, download this video and use it as input. Note: The programe is limited to process only 10,000 frames. Yolo detection does take long time and consume resources.

Objective

The main objective of the project is to detect lanes and warn driver when the vehicle drifts. As an added bonus I have added Yolo object detection that detects cars and people. Collision alert is in pipeline.

Process Steps

• Split video into frames.

• For each frame do the following pre-processing

  • Scale it down for faster processing. The image frame can be converted to grey scale as well, but for better visualization I have kept in 3 channel colors.

  • Yolo object detection. I have used a pre-trained model here in this project which does a pretty good of detecting objects and I have set the threshold to 0.5 which means it will box the objects only if the confidence level is more than 0.5.

  • Collision detection using YOLO object detection. This is achieved by using a tunneled vision in front of the vehicle to narrow down and avoid cars in other lanes being flagged.

  

  • Create area of interest (AOI) polygon aka mask.

  • Apply the mask on the frame. This helps to avoid noise and detect lanes in the AOI.

  • Apply thresholding to detect lanes.

  • Hough transformation to create right and left lanes from the thresolded image.

  • A sample frame with detected lanes and a point indicating the car center can be seen below.

  • Calculate slope of the lanes to detect drifts.

  • Alert when the drift count meets a threshold.

  • A sample frame with drift alert can be seen below .

• Stictch frames to make a outout video.

Output

Output of the processing is a video, but can made real time. For the purpose of the project output frames are stictched together to make it a video. Here is the link to the output.

Note: Output video is terrible when you view directly on sharepoint. Please download for a better quality.