Semi Autonomous Car With Self Parking
Autonomous cars are the future smart cars anticipated to be driver less, efficient and crash avoiding ideal urban car of the future. Self-driving cars are those cars in which human drivers are not required to take control to safely operate the car. That is why the self-driving cars are also known as
2025-06-28 16:29:03 - Adil Khan
Semi Autonomous Car With Self Parking
Project Area of Specialization Artificial IntelligenceProject SummaryAutonomous cars are the future smart cars anticipated to be driver less, efficient and crash avoiding ideal urban car of the future. Self-driving cars are those cars in which human drivers are not required to take control to safely operate the car. That is why the self-driving cars are also known as driverless cars. We will explore the need for an autonomous vehicle in this modern era of Science and Technology. This car is capable of moving without any human input. This car senses their surrounding using camera and ultrasonic sensor’s. Camera and ultrasonic sensor’s used to keep the track of the car position within lane. As we know that, lane detection plays an important role in the smooth flow of traffic and the identification of vehicles behind and in front of a vehicle. This will reduce the speed according to the speed of other vehicles and also detect accurate and low traffic lanes to avoid accidents. Autonomous vehicles detect and provide a way to emergency vehicles.
Layers of Autonomy: Different cars are capable of different levels of self-driving, and are often described by researchers on a scale of 0-5. Level 0: All major systems are controlled by humans.
Level 1: Certain systems, such as cruise control or automatic braking, may be controlled by the car, one at a time.
Level 2: The car offers at least two simultaneous automated functions, like acceleration and steering, but requires humans for safe operation.
Level 3: The car can manage all safety-critical functions under certain conditions, but the driver is expected to take over when alerted.
Level 4: The car is fully-autonomous in some driving scenarios, though not at all.
Level 5: The car is completely capable of self-driving in every situation.
Lane Detection: There are two type of approaches used for lane detection: the featurebased method and the model-based methods. The feature based method used to detect lanes from the capturing of images from camera and extracting low feature images. Feature based method require thousands of images of the road with well painted of lanes for proper and accurate detection of lanes. On the other hand, the model based methods use some geometrical elements to describe different type of lanes, including straight lanes, hyperbolic lanes and parabolic curves. In feature method, during the capture of images, there is also a disturbance of noise in the images. To avoid these issues, we also include model-based method.
Self Parking: Parallel parking can be challenging for drivers, including myself. The typical modern day car does not contain any systems in place to make parking easier. The main goal of this project is to design a simple prototype parking system that can perform parallel parking maneuvers autonomously. This system would include a series of proximity sensors as well as a central microprocessor that controls the car. This system would ideally work on both a scale model of a car as well as a life-sized car.
Project ObjectivesObjectives:
1. Dectection of Lanes
- Stage 1 : Lane Extraction
The boundary of the lane can be regarded as a curve, whose detection poses challenges, such as: 1. The lane markings may be occluded by shadows, tyres, etc. 2. The lane may be very short as compared to the other curves. To cater these issues, we treat the lane as a collection of several smaller line segments. Figure 3 (a) shows an edge image of the original road image, whereas Figure 3(c) consists of small line segments obtained by using our Hough Transforms. As shown in Fig. 3(b), three edge points vote to three series of points in polar space, respectively. After transformation, the collinear or almost collinear edge pixels generate stronger votes. The traditional HT determines lines according to the vote values. However, some nonexistent lines may have a high vote value due to the presence of many non-continuous collinear edge pixels. In this case, the detection of short lanes becomes highly improbable. Furthermore, it is difficult to choose an appropriate universal threshold because the same vote value may prove to be sufficient or insufficient for detection of a lane under different conditions. A low threshold will result in the detection of some nonexistent lines, while a high threshold will result in the loss of some important lines. For our modified HT algorithm, we take into account the distance between adjacent edge pixels which vote for the same line. Our modified HT relies on the notion of ‘continuity’ (given in Definition below) which enables us to detect small line segment.
- Stage 2 Small Line Segments Clusters
We cluster the set of the small line segments depending on the similarity measurement. We consider mainly two aspects:
1. The degree of collinearity measured by the radial coordinate p and angular coordinate 0.
2. The shortest distance among small line segments. Note that the higher the similarity score between segments. Cluster analysis is a common technique for statistical data analysis. Clustering algorithms can divide objects into various groups, or clusters. Similarity of objects within any cluster is maximized and similarity of objects belonging to different clusters is minimized. A number of clustering algorithms have been proposed. Among them, density-based algorithms can discover clusters of arbitrary shapes, whereby a cluster is defined as a connected dense component.
- Stage 3 Parallel Parking/ Vertical Parking:
the car crosses the parking area and the car stops when two sensors on the side see the wall again. He comes back a little and turns right 45 degrees.
If the sensors at the edges measure the value too much over the length of the car, the car stops and turns 90 degrees to the left. They start moving towards the parking lot. At this time, the front sensor continuously measures and the car stops if the measured value is less than 10cm. Park operation is completed.
Project Implementation MethodLane Detection Method:
For road lane detection we will use a Camera on the front of a car to have a view of road and then we will apply some Image Processing and Machine Learning techniques to detect the road lanes. We formulate a two stage road lane detection method to detect the boundary lines of a road lane. We consider that the middle line of a lane is represented by a curve line. In Two Stage Road Lane Detection, lane boundaries are considered as a collection of small line segments. As shown in Figure 1, S is denoted as the set of small line segments and P stands for the set of midpoints of each small line segment in S. In the ideal case, the elements of the set P should be uniformly distributed on both the boundaries of the lane, which means that any point belonging to P on one side of the boundary should have its corresponding point on the other side of the boundary. However, this is not always the case. For instance, in Figure 1, the upper part of the middle line strays towards the right boundary due to the absence of small line segments on the upper left boundary. To solve this issue, we adopt a two-stage curve fitting method. The first stage involves obtaining an initial curve for P. In the second stage, we remove some noisy points and add some new points to ensure the uniform distribution of points in P on both the boundaries.
Stage 1 (Small lines Extraction):
The boundary of the lane can be regarded as a curve, whose detection poses challenges, such as:
1. The lane markings may be occluded by shadows, tyres, etc.
2. The lane may be very short as compared to the other curves.
(a) Three edge points detected by the Canny algorithm. (b) The corresponding voting curves in polar space. Each intersection of the curves represents the line through the corresponding two points. (c) shows the results corresponding to our modified HT. (d), (e) and (f) show the results obtained by the traditional HT with a low, middle and high threshold, respectively.
Stage 2 (Small Line Segment Clusters):
As shown in Figure, each lane corresponds to many small line segments. There are some practical issues to be dealt with:
1. The offset between the actual location and the detected location of a line segment.
2. The lane boundary is not a perfect straight line, and has small curvatures.
3. Any two small line segments lying on two sides of the boundaries do not map to the same radial coordinate p.
We cluster the set of the small line segments depending on the similarity measurement. We consider mainly two aspects: 1. The degree of collinearity measured by the radial coordinate p and angular coordinate 0.
2. The shortest distance among small line segments.
Self Parking:
Case 1: If the measured value is bigger than the car and smaller than the length of the car, the parallel parking system will operate. The car comes back a little and turns right 45 degrees.
The front sensor measures and goes forward until it is small by 10 cm and stops when it is small by 10 cm. Parking is over.
Benefits of the Project:
1.Over the past few decades, use of vehicles in the world increased exponentially and this will also cause increased in road crashes which take millions of lives per year. In order to decrease deaths every year, the concept of autonomous vehicles was build. The self-driving cars are those cars in which humans are not require to take control to safely operate the cars. The aim behind the autonomous vehicle is to reduce the pollution, crashes, energy consumption and to save the human lives.
2. The lane is important of highways and Motorways as many traffic rules for controlling and guiding drivers and reduce traffic conflict over lanes, which is major part to decrease deaths. Lane detection plays an important role in the smooth flow of traffic and the identification of vehicles behind and in front of a vehicle. This will reduce the speed according to the speed of other vehicles and also detect accurate and low traffic lanes to avoid accidents.
3. Autonomous vehicles detect and provide a way to emergency vehicles.
Self Parking:
Parallel parking can be challenging for drivers, including myself. The typical modern day car does not contain any systems in place to make parking easier. The main goal of this project is to design a simple prototype parking system that can perform parallel parking maneuvers autonomously. This system would include a series of proximity sensors as well as a central microprocessor that controls the car. This system would ideally work on both a scale model of a car as well as a life-sized car. This project explores how sensor input and an algorithm can be used for practical applications.
1.As the population increased in the cities, the usage of vehicles got increased. It causes problem for parking which leads to traffic congestion, driver frustration, and air pollution. When we visit the various public places like shopping malls, multiplex cinema hall and hotels during the festival time or weekends it creates more parking problem. In the recent research found that a driver takes nearly 8 minutes to park his vehicle because here spends more time in searching the parking lot. The searching leads to 30 to 40 percent to traffic congestion. Here we going to see how to reduce the parking problem automatic car parking using offerings are transforming cities by improving infrastructure, creating more efficient and cost effective municipal services, enhancing public transportation, reducing traffic congestion, and keeping citizens safe and more engaged in the community.
Technical Details of Final DeliverableFinal Lane Detection Completed :
We have obtained the set of clusters with the help of the straight line feature of the lane boundary. However, such feature is not enough to distinguish lanes from other objects. There are some noise clusters which have strong features of the straight line. Therefore, we identify the final lanes with two strategies: 1. Find the candidate clusters which have high probability corresponding to a lane using a special characteristic of the lane. 2. Identify the final lanes combining with the vanishing point. We employ a near exhaustive strategy in the candidate clusters because the number of the clusters is not much. Taking the left lane as an example, we get all possible curve lines for different combinations of clusters. We deal with the right lane in the same way. Considering the lanes are parallel, therefore, the lanes and vanishing point should satisfy the condition: Identification of Lanes 1
The lane should pass through the vanishing point.
2. a1a2 and B should tend toward zero. Here, a1 and a2 are angles between the corresponding lanes respectively. B is the angle between the lines and the vanishing point. (a) shows the direction of small line segments. dx and dy represent the horizontal and vertical gradients of the edge pixel, respectively. rl and rr are the left and the right boundaries of the right lane, respectively. (b) shows the condition what vanishing point and the lanes should satisfy.
Self Parking:
The Decision model that begins the parking procedures are based on three premises depending upon the available parking space or distance between 10 cars,
1. If parking space is larger than car, forward parking movements are performed.
2. If parking space is enough, backward parking movements are performed.
3. But if parking space is shorter than car, no action is performed. Hence, parking space is one of five measurements inputs in the system which helps the system to decide whether it is possible to park the car or not. The other four measurements correspond to distance from side walk, distance from front car, distance from back car, and inclination (Inclination). Once system has acquired the parking system value and it is in the permissible parking range, the car must take an initial position so that the scaled car performs movements to park and it stops until it satisfy the desirable parking conditions of rules. Then the fuzzy model which decides which action has to be taken at the beginning of the process is named decision model. It determines whether backward model or forward model has to be performer after parking space has been measured. One has to understand actions as the movements that an autonomous control has to perform instead a human driver. And these actions are in function of the controlling system outputs. The controlling car outputs are the following:
1. Car direction or tires angle.
2. Speed of Car.
3. Direction of movement, Forwards or Backwards.
Final Deliverable of the Project Hardware SystemCore Industry TransportationOther Industries Manufacturing , Others Core Technology Artificial Intelligence(AI)Other Technologies Internet of Things (IoT), Robotics, NeuroTechSustainable Development Goals Affordable and Clean Energy, Industry, Innovation and Infrastructure, Climate ActionRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 46350 | |||
| Jetson Nano 2GB | Equipment | 1 | 15500 | 15500 |
| Arduino Uno | Equipment | 1 | 1000 | 1000 |
| Arduino Nano | Equipment | 2 | 1200 | 2400 |
| Ultrasonic sensor | Equipment | 4 | 300 | 1200 |
| Motor driver | Equipment | 2 | 400 | 800 |
| Servo Motors | Equipment | 2 | 400 | 800 |
| DC motors | Equipment | 2 | 350 | 700 |
| Prototype car | Equipment | 2 | 1500 | 3000 |
| Servo and DC driver | Equipment | 1 | 1500 | 1500 |
| Camera V2 | Equipment | 1 | 5500 | 5500 |
| Memory Card | Equipment | 1 | 2000 | 2000 |
| Transmitter and receiver | Equipment | 2 | 700 | 1400 |
| Power for Jetson | Equipment | 1 | 2500 | 2500 |
| External DC supply ( lithium Cell) | Equipment | 9 | 300 | 2700 |
| ICs ( Different) | Equipment | 10 | 100 | 1000 |
| Miscellaneous items ( wires, soldering Iron, DMM etc) | Miscellaneous | 1 | 4000 | 4000 |
| Boost converter | Equipment | 1 | 350 | 350 |