Abstract
The intervention of AI technology and self-driving vehicles changed transportation systems. The current self-driving vehicles demand reliable and accurate information from various functional modules. One of the major modules accommodated in vehicles is object detection and classification. In this paper, a speed bump detection approach is developed for slow-moving electric vehicle platforms. The developed system uses monocular images as input and segment the speed bump using the GAN network. The results obtained by the new approach show that the GAN network is capable of segmenting various types of speed bumps with good accuracy. This new alternative approach shows the ability of GANs for speed bump detection applications in self-driving vehicles.
-
Author Keywords
- Conditional Generative Adversarial Network (cGAN) ,
- Self-driving vehicle,
- Speed bump detection
-
Controlled Indexing
- electric vehicles,
- image classification,
- image segmentation,
- intelligent transportation systems,
- neural nets,
- object detection,
- road vehicles
-
Non-Controlled Indexing
- object detection,
- object classification,
- slow-moving electric vehicle platform,
- GAN network,
- speed bump detection application,
- speed bump segmentation,
- conditional generative adversarial network,
- transportation systems,
- self-driving vehicles,
- functional modules,
- AI technology,
- monocular images
Introduction
Today with the advancement in computer vision technologies, self-driving vehicles are booming in the market. 2019 Autonomous Vehicles Readiness Index (A VRI) provides the fact that countries from all over the world are making progress towards a future with autonomous vehicles [1]. It indicates that governments are focusing on advanced transportation systems and taking steps that consider the real-life impact of driverless cars. Self –driving cars are going to change the driving pattern of the vehicle, where traditionally the vehicles were supposed to be driven by humans as they are capable of taking the right decisions in a dynamic environment. The modern powerful AI algorithms are challenging t h e human decision io n mak in g cap ab it is for driving a vehicle [2]. However, with the current development in AI, ML algorithms are not robust enough to replace humans because machines miserably fail for the untrained s scenarios. Th ere are s t ill s o me o p en challenges for self-driving vehicles like object detection, localization, scene prediction, vehicle control, and decision making [3]. The major challenge of self-driving vehicles is to be able to sense the surroundings and make feasible decisions. This is being obtained by advanced computer vision technologies equipped with AI algorithms. The vision sensor plays a major role in functionalities like navigation, object detection, and classification [4]. The approach used in this paper exploits the vision and neural networks to address the speed bump segmentation using a monocular camera. Detection of a speed bump at the right distance plays an important role to avoid damage to passengers and the vehicle suspension systems. As a result, prior knowledge of the speed bump can help to warn the driver or to take control action to reduce vehicle speed.
The speed bump detection was the major challenge from year’s back and various approaches were developed to address this issue. Some earlier approaches for speed bump detection uses conventional image processing techniques. First, the image is preprocessed to enhance the features, and then structural information is recovered in order to decide the presence of a speed bump [5], [6]. The above-mentioned method addresses marked and unmarked speed bump detection but fails when the image captured under different lighting conditions. Apart from this, other methods use special sensors like depth, GPS, and accelerometer [7], [8], [9]. The hardware-based solutions are costly and they are more prone to errors and have limitations over the range, therefore real-time usage of such implementation is difficult. Using a vision-based approach for this problem will not only reduce the need for special-purpose hardware but also will reduce the cost of the system. The deep neural networks (DNNs) can be used to address vision-based problems, recent development in deep neural networks shows very good accuracy for object detection[10], [11]. The accuracy of DNNs is improved by training the network with huge data which is a major requirement for applying any DNN an algorithm. Co collecting enormous training data is a very complicated and time-consuming task, here comes the need for a DNN algorithm that can perform well even with less training data.
In this paper we are exploring an application of DNN based model called generative adversarial network (GAN) for segmentation of speed bump [12]. GAN comprises of two networks a discriminator and a generator. The generator tries to fool the discriminator by generating false data samples. Contrarily, the discriminator attempts to distinguish between the false and real data samples. Both the networks work in counter to each other during the training phase. The training process is continued until the discriminator finds it difficult to distinguish between generated fake samples and real samples [12]. There are different implementations of GAN depending on their applications like vanilla GA N, deep convolutional GA N(dcGAN), Laplacian pyramid GA N(lap GAN), super-resolution GAN(srGAN), and conditional GAN (cGA N) [12], [13], [14], [15] [16]. cGAN is very popular and it is being used for a variety of image generation and manipulation applicant io n s. Ph Phillip Is o la et al. proposed Image -to -Image translation method called pix2pixusing cGAN. It serves a common platform for all images to image translation problems rather than using an individual approach for each application. Experiments conducted using their proposed approach include labeled images to photo, Map to the aerial photo, Black and White to color photos, day to night, etc. The same cGAN is able to perform all the above-mentioned images to image translation with better efficiency [17]. The proposed method models problem as an image to image translation for segmenting the speed bump from the input image.
Conclusion
We have achieved marked and unmarked speed bump segmentation using Conditional GANs. Fro m the results obtained above we can conclude that GANs can be used as another alternative approach specifically where we suffer from a lack of training data for applying deep learning algorithms. The proposed methodology illustrates the ability of conditional GAN as an application of speed bump detection for self-driving cars. The CGA N implementation presented in this paper gives us a promising alternative approach for the speed bump segmentation. From the experiments conducted it is evident that the network performance is good even with a small training dataset. The major contribution of the future work will be to include the depth information, distance estimation, and increasing the accuracy of segmentation.
About KSRA
The Kavian Scientific Research Association (KSRA) is a non-profit research organization to provide research / educational services in December 2013. The members of the community had formed a virtual group on the Viber social network. The core of the Kavian Scientific Association was formed with these members as founders. These individuals, led by Professor Siavosh Kaviani, decided to launch a scientific / research association with an emphasis on education.
KSRA research association, as a non-profit research firm, is committed to providing research services in the field of knowledge. The main beneficiaries of this association are public or private knowledge-based companies, students, researchers, researchers, professors, universities, and industrial and semi-industrial centers around the world.
Our main services Based on Education for all Spectrum people in the world. We want to make an integration between researches and educations. We believe education is the main right of Human beings. So our services should be concentrated on inclusive education.
The KSRA team partners with local under-served communities around the world to improve the access to and quality of knowledge based on education, amplify and augment learning programs where they exist, and create new opportunities for e-learning where traditional education systems are lacking or non-existent.
FULL Paper PDF file:
Speed Bump Segmentation an Application of Conditional Generative Adversarial Network for Self-driving VehiclesBibliography
author
Year
2020
Title
Speed Bump Segmentation an Application of Conditional Generative Adversarial Network for Self-driving Vehicles
Publish in
2020 Fourth International Conference on Computing Methodologies and Communication (ICCMC), Erode, India, 2020, pp. 935-939,
Doi
10.1109/ICCMC48092.2020.ICCMC-000173.
PDF reference and original file: Click here
Somayeh Nosrati was born in 1982 in Tehran. She holds a Master's degree in artificial intelligence from Khatam University of Tehran.
Professor Siavosh Kaviani was born in 1961 in Tehran. He had a professorship. He holds a Ph.D. in Software Engineering from the QL University of Software Development Methodology and an honorary Ph.D. from the University of Chelsea.
Nasim Gazerani was born in 1983 in Arak. She holds a Master's degree in Software Engineering from UM University of Malaysia.
