Abstract: We start with the premise that the combination of different computing technologies embedded in self-governing cars is a powerful tool for efficiency in communications, information collection, processing, and storage. However, by focusing on efficiency, self-driving vehicles provide a new mode of industrialized transportation whose users can only choose between transportation services but have little or no say about the broader social implications of the technology.Since 81 percent of car smashes are the consequences of human inefficiency, Self-governing cars would take a lot of danger out of the equation entirely. Many cars are already equipped with features in the first stage of “automatic” driving, like autonomous braking, self-parking, or sensors that clue a driver into a nearby obstacle.Using a microcomputer that is the raspberry pi as the Electronically Controlled Unit(ECU) for the vehicle and using deep learning for lane following and sign detection. We will train the model on AWS, and run the trained model on a localhost.The ECU will send the input it observes to the localhost. The localhost responds with the actions to be taken and followed.General Terms:Artificial IntelligenceComputer VisionMachine Learning methods Computer systems organizationRoboticsKeywords:Scene understanding Neural networksVisual content-based indexing and retrieval Vision for robotics Object detectionRaspberry Pi 3 Model B Introduction:1.1 BACKGROUNDIndia, being the epicenter of population boom and an increasing technological hub with centers like Pune, Banglore we nd that the over increasing population and the resulting increase in accidents have become one of the major problems to solve.We nd that most of the accidents are happening due to human negligence and haste to reach work on time. Most of the cities are now being connected by freeways and have increased the possibilities of accidents as vehicles are speeding over the mentioned speed limit. Here, the Self-Driving Vehicle, if used on highways would reduce the accidents happening significantly. Our self-driving vehicle would follow the lanes and the speed limit boards mentioned along with traffic signals.Even though there are pioneers like Tesla Motors and NVIDIA for self-driving vehicles, even if a standard level of autonomy like lane following, sign detection, obstacle detection and signal detection implemented in the vehicle would reduce casualties significantly. 1.2 PROBLEM STATEMENTPeople with impairment face many complications. These people aren’t able to commute to places with being dependent on others.It has raised the topic of the self-governing car and their applications. Such people can be less dependent and can commute to places efficiently. Additionally, The human motorist is the main reason for on-road casualties. whereas, the mishaps by self-governing car are negligible. The self-governing cars are an actually better motorist as compared to the various bad etiquettes a driver of the car might exhibit.Since 81 percent of car crashes are the result of human inaccuracy, the autonomous car would take a lot of danger out of the equation entirely. Many cars are already equipped with features in the first stage of “automatic” driving, like self-parking, autonomous braking, or sensors that indicate a driver into a proximate hindrance. The casualties caused due to driving under the influence of alcohol should decrease because there’s no designated driver required when the car drives itself. Hence, the use of self-governing vehicles provides us a lot of boons. 1.3 PURPOSEThe purpose of the project is to successfully build a self-driving vehicle suited for Indian habitat. 1.4 SCOPEAutonomous vehicles which use deep learning promises improvement of identifying areas in which it focuses interventions as well as improves the way of implementing it. Thus, helping Indian citizens and to reach their destinations safely and reducing traffic congestion. 2. Neural Network Neural networks are flexible, nonparametric modeling tools. They can perform any complex function mapping with the desired accuracy. An ANN is typically composed of several layers of many computing elements called nodes. Each node receives an input signal from other nodes or external inputs and after processing the signals locally through a transfer function, it outputs a transformed signal to other nodes or final result. Input Layer – The input layer gets the values from a vector of the predictor variable. At the input layer, the values are distributed to each of the neurons in the hidden layer. In extension to the predictor variables, there is a constant input named the bias that is given to each of the hidden layers. The weights are multiplied by bias and given it to the sum going into the neurons. Hidden Layer – At the hidden layer, weight is multiplied by the neuron’s input value, and the resulting weighted values are added together producing a combined value. The weighted sum is fed into a transfer function. The outputs from the hidden layer are distributed to the output layer. Output Layer – On returning at the output layer, The weights are multiplied by each hidden layer value and the arising weighted values are added together producing a combined value. Thereafter, the weighted sum is fed into a transfer function, which outputs its own value.