BIO

Autonomous Toy Car using Computer Vision and Deep learning techniques.

In this project, a toy car is converted into a self-driving car capable to move autonomously under a constrained environment. Self-driving capability is developed using computer vision and deep learning techniques. Initially, tracks are deployed on the surface in order to gather data through video streaming by using a webcam. From these video, images are extracted for classification of data into four different classes, i.e; right, left, forward and stop. This classified data is converted into required format using computer vision algorithms such that the data consists of only bright Hough lines on a black image. This data is then trained using Convolutional Neural Network such that the trained model is able to predict in real-time whether to move the steering of the car left, right, forward or stop, accordingly. This project does not include any sensor and is totally computer vision based. The training and inference is done using a core i5 laptop. The trained model takes the input images from live feed and predicts which direction to choose or stop. The trained model after prediction generates a string and through serial communication the string is sent to Arduino. Finally, the Arduino processes the wrappers embedded in its code according to the string received from the trained model and sends control signals to the motors of the toy car to move or stop according to the prediction.