Automated Obstacle Detection using CNN
Data science projects in pondicherry
Create New

Automated Obstacle Detection using CNN

Project period

07/06/2021 - 09/29/2021




Project Category


Automated Obstacle Detection using CNN
Automated Obstacle Detection using CNN

A self-driving automobile or an autonomous automobile / driverless automobile by definition could be a vehicle that's capable of sensing its setting and moving with very little or no human input . During this primarily based (single input) automobile, it'll get its input from the pi camera. Raspberry Pi has been used as a process unit during this project, that takes the input through the raspberry pi camera and drives the automobile by causing management signals to the L298 2A driver module, that successively controls the automobile. The core of the project is the Convolutional Neural Network that maps the raw image pixels into options that sight objects and navigate the automobile. The self-driving automobile is one among the most effective inventions created by grouping and would relieve several issues like driving. once you square measure sleepyheaded or slopped. it might improve safety throughout the night drives that could be a necessary want within the current, the struggle for a parking spot is resolved and because it might park on its own and obtain the owner where he's. It caters to the transportation desires of individuals of all ages. All self-driving cars square measure identical provided they're on identical level. Driverless cars square measure classified on the premise of their ability to operate mechanically and square measure classified beneath totally different levels. There square measure five totally different levels, ranging from LEVEL1 wherever the motive force needs to have his active, which suggests that driver help is necessary. In LEVEL2, the motive force will have his Hands off because it could be a type of partial automation wherever the technology will at the same time management steering and speed at identical time, while not driver intervention for brief periods. LEVEL3 type of automation allows the motive force to own his Eyes off because the vehicle is capable of taking full management and operational throughout choose elements of a journey once sure operational conditions square measure met. The LEVEL4 or the high automation technology does not want an individual's driver which means that he will take his Mind off. The vehicle will basically do all the driving, however the motive force will intervene and take hold pro re nata. Finally, the last level being LEVEL5 or the complete automation could be a technology which might pay attention of the driving state of affairs all told extreme conditions and therefore the humans square measure simply passengers

Why: Problem statement

In order to extend road safety autonomous vehicles square measure beneath development and square measure the main target of the many analysis comes. The Rochester Institute of Technology desires to possess a re­entry purpose to the sector of autonomous vehicle analysis and presently doesnt have a vehicle to use as a start line. To facilitate RIT’s goal a far off management golf cart decided to be the primary stepping stone. The scope of this project includes changing a golf cart to a far off management vehicle. the security of the vehicle’s passengers and bystanders is of the utmost concern. thus the vehicle is needed to be low speed and contain the power for passengers to require management.

How: Solution description

A miniature version of a self-driving automotive has been enforced, that is capable of higher cognitive process with the assistance of Convolutional Neural Networks. The automotive was trained on a self created track that was around twelve to fourteen meters long. throughout the coaching section, the automotive had achieved associate degree accuracy of eighty eight.6%, which might be improved additional by hyperparameter standardisation and a lot of information assortment. The automotive functioned o.k. for many of its journey however had difficulties once taking sharp turns, that was because of the limitation of the motor’s turning capabilities. The automotive had with success mimicked the driving patterns at intervals the scope of the informationset and would need a lot of data accumulation for adjusting to new eventualities.

The automaton are ready to move as per the command given once detective work the obstacle through the camera module. once the video frame containing the obstacle is detected exploitation the image process algorithmic rule the pi can command the motor as per the directions i.e. left or right and till amendment its path consequently.

Fig1: Schematic diagram of the obstacle detection car


Fig2: Block diagram of Obstacle detection car
Output Screenshot:

How is it different from competition

This section describes the small print of implementation once the input pictures were recognized as either Boom Barrier, Plastic Barricade,Flap Barrier,reflective traffic cone, PVC cone and human,animals etc. we have a tendency to assumed that each one varieties of obstacles were recognized in a picture.

Who are your customers

  • Obstacle avoiding automatons is employed in the majority mobile robot navigation systems.
  • They can be used for unit work like automatic vacuum improvement.
  • They can even be employed in dangerous environments, wherever human penetration might be fatal.

Project Phases and Schedule

Phase 1: Collect Images

Phase 2: Build an AI model and train objects

Phase 3: Visualization and result

Phase 4: Fix the hardware component

Phase 5: Interface the rover command line to the image processing method

Resources Required

 List of hardware:

  • Raspberry Pi 3 model B
  • Motor drivers(L298N)
  • Wi-Fi 802.11n dongle to connect remotely to pi.
  • 8AAA batteries
  • DC motors
  • Web camera

A. Hardware and software package

The automotive framework was engineered employing a stripped RC automotive, we have a tendency to unbroken the most frame and also the 2 DC motors of the RC automotive. associate degree L298 2A motor driver was accustomed management the 2 DC motors. L298 2A contains 2 full H-bridges that change North American nation to regulate 2 DC motors bi-directionally, one for the acceleration and another for steering movement of the automotive. All the connections were done as mentioned. The Raspberry Pi may be a tiny however full-featured pc on one board, this is often a sensible, moveable and cheap device. we have a tendency to used Raspberry pi 3B for process still as for dominant the automotive. A raspberry camera is connected at the front of the car; it'll collect the image information throughout the coaching method.

B. information collection: For Building a Neural network model we want various information. the information is

collected throughout the coaching method. Initially, the automotive has to be controlled wirelessly. to try to to thus we have a tendency to North American nationed VNC viewer that permits us to regulate Raspberry Pi wirelessly through Wi-fi in devices like laptops and smartphones. Next is to make a track so we will train the automotive. As we have a tendency to run the automotive on the track it collects the image information through the raspberry pi camera and also the input command (forward, stop, right, left). This information are accustomed train the neural network model.

C. Convolution neural network:

Convolutional neural networks(CNN) became omnipresent in pc vision. CNN is hottest as a result of reliable results on visual perception and detection that square measure helpful in real-world applications. the everyday use of convolutional networks is on classification tasks, wherever the output to a picture may be a single category label that has to be classified. Convolution may be a perform derived from 2 given functions by integration that expresses however the form of 1 is changed by the opposite. in contrast to neural networks, the input to CNN is a picture. The convolution layer contains a collection of freelance Filters. The filter is softened over the entire image and also the real number is taken between the filter and chunks of the input image. every filter is severally convolved with the image and finally ends up with feature maps. There square measure many uses that we have a tendency to gain from derivation a feature map, reducing the dimensions of the image by protective it’s linguistics data is one amongst them.

View on Github
Automated Obstacle detection using CNN


Leave a Comment

Post a Comment

Are you Interested in this project?

Do you need help with a similar project? We can guide you. Please Click the Contact Us button.

Contact Us

Social Sharing