Driver Drowsiness Detection Using Deep Convolutional Neural Network
Thousands of accident take place worldwide every year due to drowsiness. So it is need of time to develop a drowsiness detection system that can try to overcome these accidents due to drowsiness. In this work, we will use a deep learning methodology which will be based on convolutional neural networ
2025-06-28 16:32:12 - Adil Khan
Driver Drowsiness Detection Using Deep Convolutional Neural Network
Project Area of Specialization Software EngineeringProject SummaryThousands of accident take place worldwide every year due to drowsiness. So it is need of time to develop a drowsiness detection system that can try to overcome these accidents due to drowsiness. In this work, we will use a deep learning methodology which will be based on convolutional neural networks (CNN) to solve this problem. This methodology will perform drowsiness detection from an incoming video stream of a driver, detects and localizes open and closed eyes as an object detection task. Raspberry Pi architecture with Single Shot Multi Box Detector (SSD) is used for object detection task. Then an algorithm will be used to detect drivers drowsiness based upon the output from Raspberry Pi–SSD architecture. In order to train this architecture a custom data set of images will be compiled and used to test the trained model. The proposed methodology will be reasonably more accurate, more computationally efficient and more cost effective as it can process the input video in a cheap Raspberry Pi device.
Project ObjectivesThe objectives of the project is to:
1) Study hardware specific optimizations like overclocking, GPU, increasing swap and GPU memory, multi-threading, etc.
2) Study software specific optimizations like small filter size, small number of filters, 1x1 convolutions for dimensionality reduction, depthwise separable convolution, etc. that are utilized in the design of lightweight CNNs
3) Implement deep learning based accurate and real-time driver drowsiness detection system using software and hardware optimizations on embedded hardware like Raspberry Pi 3B+.
Project Implementation MethodThis project proposes a deep convolutional neural network (CNN) architecture for detection of driver drowsiness and deployment on Raspberry Pi. The purposed approach will input the videos frames and will consider the eyelid open and closure duration blink frequency and yawn as signature indicator. Previously purposed techniques performance is affected by low luminance condition, poor contrast and varying user appearance. More over most of the detection system involve a feature extraction step, which is accounted as time consuming task and reduce applicability in real time detection. The purposed approach will address the limitation of previously purposed approaches and image processing related problems by using CNN which can automatically extract features from raw images and videos and has shown exemplary performance. Deep CNN architectures are heavy and computationally intensive. In this project deep CNN architecture will be designed and trained by exploiting transfer learning drowsiness detection problem will be consider as object detection problem therefore Single Shot Multi box Detector (SSD) algorithm will be modified and optimized for detection of indicator feature the optimized architecture will be squeezed and deploy on Raspberry Pi. The application will alert the drowsy driver by alarm the performance of detection application in Raspberry Pi and will be evaluated in real time driving condition.
Benefits of the ProjectVehicle crashes and accident due to drowsy driving are prevalent all over the world. Thousands of people die every year resulting from vehicle accident due drowsy driving. The major reason for these accidents is driver fatigue or sleepiness. In order to reduce such accidents and to enhance the safety of the driver and the passengers, drowsiness detection system can be used.
Technical Details of Final Deliverable•Drowsiness detection problem will be consider as object detection problem therefore Single Shot Multi box Detector (SSD) algorithm will be modified and optimized for detection of indicator feature the optimized architecture will be squeezed and deploy on Raspberry Pi.
• The application will alert the drowsy driver by alarm. the performance of detection application in Raspberry Pi will be evaluated in real time driving condition.
Final Deliverable of the Project HW/SW integrated systemCore Industry TransportationOther Industries Others Core Technology Internet of Things (IoT)Other Technologies OthersSustainable Development Goals Good Health and Well-Being for People, Industry, Innovation and InfrastructureRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 31100 | |||
| Raspberry Pi Model 3B+ | Equipment | 1 | 6000 | 6000 |
| Raspberry Pi IR Camera 8 MP | Equipment | 1 | 5000 | 5000 |
| keyboard | Equipment | 1 | 1200 | 1200 |
| memory card 32/64 gb | Equipment | 1 | 2000 | 2000 |
| Lcd Screen | Equipment | 1 | 8000 | 8000 |
| Mouse | Equipment | 1 | 500 | 500 |
| Hdmi connector | Equipment | 1 | 800 | 800 |
| Buzzer | Equipment | 1 | 100 | 100 |
| Thesis documnetation | Miscellaneous | 3 | 2500 | 7500 |