Autonomous vehicle using Raspberry with Intel Movidius™ Neural Compute Stick.
- 0 Collaborators
AI is moving towards Edge(standalone offline devices) rather than cloud as the latency in the upload of raw data and download of processed data is quite high with respect to the real-time application like self-driving cars. An example is Intel Movidius™ Neural Compute Stick. It allows offline processing of images with deep-learning i.e Computer Vission. The objective of this project is to exploit the capabilities of the NCS for real-time computer vision applications ...learn more
Project status: Under Development
Robotics, Artificial Intelligence
Groups
Student Developers for AI,
DeepLearning
Intel Technologies
Intel Opt ML/DL Framework,
Movidius NCS
Overview / Usage
Object detection is the most important task of the Self-driving car operation pipeline. The detection should be accurate as well as fast. There are various techniques like YOLO( You Only Look Once) and SSD(Single Shot Detection) which are accurate and fast but need high-end GPUs for real-time operation. In this project, either YOLO or SSD will be implemented using tensorflow on the Intel NCS. A Raspberry PI will be used used to control the autonomous vehicle. The input will be either Raspberry Pi camera or a HD webcam. The vehicle will run on a dummy track with signals and obstacles like other vehicles and pedestrians.
Methodology / Approach
A deep neural network will be built for the NCS using NCSDK2 based on the Tensorflow Framework. After appropriate training and testing, it will be deployed on Raspberry Pi(RPi). The RPi will be connected to motors to control the vehicle. The vehicle will stop in case a red traffic light is detected will slow down n case any obstacle comes in front.
Technologies Used
Intel Movidius™ Neural Compute Stick
NCSDK2