Description:

The current invention consists of a cost-effective and energy efficient system, wherein a driverless vehicle is capable of interacting with its surrounding BLE devices and responding to nearby traffic entities in real-time. The car has a single wide-angle camera with which it can perceive its surroundings and thereafter measure distances of incoming or static obstacles, triggering an appropriate response.

BLE or Bluetooth Low Energy Technology is focused to provide reduced power consumption, parallely, maintaining it’s range as a replacement for the GSM module. It, combined with a NRF24L01 transceiver, constitutes a Beacon, which can be installed in a locality, centrally or in clusters. Beacon is a device that produces a single signal, that is visible to all those devices in it’s range. Several Beacon checkpoints have been placed in the roadmap for the car to interact with, while travelling.

The car provides three different modes of communication with the user, namely: ‘Internet Mode’, ‘Bluetooth Mode’ And ‘SMS Mode’.The car will switch to a particular mode as per convenience, considering network availability, signal strength and scope of resources.

Considering the “SMS Mode” or “Short Messaging Service Mode”, the car will have a GSM Module interfaced to it, so as to interact with the user. A GSM module(Global System for Mobile) is used to establish communication between a mobile device or a computing system and a GSM system. Breaking the stereotype, of always being dependent on Internet, we use GSM. The user won’t be required to have an active internet connection, as he can just interact with the car, who can now receive messages with the help of GSM.
When the user wants to summon the car, his GPS location will be sent, over a GSM network, to the car. The car will receive the location and proceed accordingly.

When proper network connectivity is available, the “Internet Mode” comes into play. The user may send the request and also his GPS location, using a mobile application via internet which is stored in a web server, and correspondingly received by the car. Using the GPS locations of the car and the user, the nearest Beacon checkpoints are determined. Once the initial and the destination checkpoints are noted, the car advances by interacting with the intermediate Beacon checkpoints, thereby driving through the determined path.

Under circumstances when cellular network connectivity is not reliable, the user can connect to the car using the “Bluetooth Mode”. One can use their mobile devices to send a summoning request via bluetooth. The BLE transceivers or Beacons are placed throughout the local roadmap at strategic locations. These transceivers continuously emits signals which are identified by the car. The beacons help to relay the request signals from receiver end to the sender end paired devices, via bluetooth. The user can be at any location and send out a summoning request signal conveniently with the help of a mobile application, to the car. The car, having a Beacon interfaced with it, will receive the command and try to receive signals from the beacons (on the transceivers) in between the user and itself. Once it has received the signals, it’ll retrieve the user's location and figure out a possible path to the user.

For this, the car has to previously store the data of the respective surrounding local roadmap. As previously mentioned, the road network has multiple transceivers which emit/receive signals at regular intervals. The car determines the known location closest to the user and arrives there.

The roadmap has been represented in form of a bidirectional graph with each vertex representing a Beacon placed in the locality. The edges of the graph are weighted, each containing a numeric value between 0 to 359, representing the orientation of the path from one vertex to other, in form of degrees from North direction. The current direction from North, is measured by a Magnetometer, attached in the car. It then correspondingly aligns itself to the correct direction and starts moving forward.

While moving, the car will constantly keep interacting with the neighbouring Beacons, by observing their signal strength, as a function of distance between them, so as to reach the appropriate location, step by step. On reaching the final Beacon checkpoint, the car will refer to the determined location of the User and finally orient itself and move towards the intended destination.

During driving, the car is using Haar Cascade classifiers implemented in OpenCV. It is used for detecting obstacles, road signs, other vehicles etc. The image data collected are turned grayscale, resized accordingly and then processed. We have deliberately used a haar-cascade classifier to reduce the computational complexity as compared to processing pixel values which would have made the detection too slow and hence very impractical.

The classifier of Haar-like features goes through a list of stages where each stage in turn is a list of weak learners. It works by moving a window over an image to identify a particular object. The region in which the object is identified is labeled by the current location of the window as either positive or negative - positive meaning that an object was found in a particular image and negative meaning that it was not. If the labels give a positive result, then the classifier moves onto the next stage. If not, the window used for identification moves onto the next location. The classifier gives a final answer of positive when all the stages say that the object is found in the image.

Another way to detect objects would be using HOG or the ‘Histogram of Oriented Gradients’. It is a feature identifier used in computer vision and image processing for the purpose of object detection. The approach keeps a track of the occurrences of inclined orientation in confined sections of an image.

The above method distinguishes the boundaries, inclined structures and regional shapes in a local portrayal with easily governable degree of invariance to local photometric and geometric conversions. These variations could incorporate adaptations or rotation, which do not cause any considerable change if they are smaller than the regular structural or direction bin size.

We have used a Convolutional Neural Network to implement our decision-making model. Neural networks are a state of an art technique to solve various problems from predicting housing prices to making a chatbot. The best thing about neural networks is that they don’t require any hand coded feature engineering. The network learns to pick which input features to choose and select.Neural network can be divided into three layers. They are input layer, hidden layer and output layer. The input layer consists of input features in form of vector. The hidden layer consist of multiple layers each containing neurons. These layers can go very deep and the number of neurons can range from several hundred to thousands.

Regular neural networks don’t perform well on image dataset as the input parameters is a 3D matrix which can have large size , so to solve this problem Convolutional Neural Networks which reduces the matrix into smaller size of parameters which could me flattened and fed into a regular neural network. Convolutional neural network are used for building powerful image recognition and object detection systems. Convolutional neural network takes input as image which is a 3D matrix consisting of height,width and channels of the image. For a grayscale image the channel is one and for a color image the channel taken is three for RGB values.

The first operation in Convolutional neural network is convolution which consists of filters that can be learned. Each of these filters detect different things such as edges,lines,etc. The dot product of the input matrix and filters are performed.Padding is used to extract more features from edges and corners of the image. Batch Normalization technique is used to speed up learning. Then ReLU is applied as an activation function and also to increase the non-linearity of the matrix. After that pooling operation is performed which down samples the matrix and generalizes the matrix from the convolution operation.The two most commonly used pooling operation are max pooling and average pooling. After several operations of convolutions and pooling the 3-D matrix is finally flattened into a 1-D array. This array is fed to fully connected neural network. At the end of this fully connected neural network, the output layer is present which contains one or several neurons. While training the network data augmentation is used to reduce overfitting. Another techniqnique to reduce overfitting is dropout regularization that randomly drops some neurons in a particular layer while training.

The model we will use will take the input image of the road and that will be passed to a convolutional neural network consisting of several convolution and pooling operation. At the end of the fully connected network in the output layer there will four neurons each representing directions i.e :- forward,right,left and stop. Many techniques mentioned above like batch normalization, data augmentation, etc will be used to improve the accuracy of the network.

Thus, this system of communication for driverless cars can be used anywhere; underground parking stations, or inside the house or outside on the road. Thus our model of the driverless car improves its functionality and also increases its applications.

We claim:

1. A system for driverless cars to interact with its surroundings through Bluetooth Low Energy(BLE) Technology consisting:

* A BLE communication device mountable to wayside objects like electric poles.

* A transmitting module placed in the BLE communication device to constantly receive or emit bluetooth signals.

* A receiving module placed in the vehicle to receive signals from the above module.

2. The BLE communication system, according to claim 1, wherein the BLE devices interact with the driverless car only using Bluetooth technology throughout the drive, eliminates the necessity of the vehicle being equipped with continuous GPS or internet capabilities.

3. The BLE communication system, in reference to claim 2, wherein a standard routine of autonomous switching between Global System for Mobiles(GSM) and Bluetooth Low Energy technologies is implemented to receive the location information of the person summoning the vehicle, as and when situation demands.

4. The autonomous switching between GSM and BLE, in context to claim 3, wherein the best suitable technology is determined by the availability of network and BLE signal strength.

5. The communication system, according to claim 3, wherein provision for text based service requests is made to enable legacy mobile devices incapable of internet facilities to send a summon request to the driverless car.

6. The BLE communication system, in reference to claim 2, wherein the driverless vehicle need not be equipped with continuous GPS and internet capabilities once the destination is set, thereby making the technology available to a larger population and developing areas.

7. The BLE communication system, according to claim 2, wherein the receiver module in the car only interacts with local BLE devices in its immediate surroundings, rendering the system independent of weather anomalies and background noises, in one or more modes of operation.

8. The communication system according to claim 1, comprising a mobile application to communicate with the driverless car, capable of working over both Bluetooth and Internet.

9. A system of determining an appropriate path and direction between initial and final locations by a driverless vehicle consisting of:

* A magnetometer to perceive real-time orientation of the vehicle,

* A receiving module placed in the vehicle to receive signals from the surrounding BLE transmitter modules.

10. The path determination system, according to claim 9, wherein an efficient representation of local roadmap comprising both location and direction data.