Akarsh Singh
Akarsh Singh has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Amol Dhumane
Symbiosis Institute of Technology, Pune Symbiosis International Deemed University, Pune, Maharashtra, 412115, India
1 shared publication - Ayush Koul
Symbiosis Institute of Technology
1 shared publication - Maruti Panchal
Symbiosis Institute of Technology
1 shared publication - Natasha Jhamnani
Symbiosis International (Deemed)University (SIU), Lavale, Symbiosis Institute of Technology, India
1 shared publication - Om Hotkar
Symbiosis Institute of Technology
1 shared publication - Pratheeik Radhakrishnan
Symbiosis International (Deemed)University (SIU), Lavale, Symbiosis Institute of Technology, India
1 shared publication
Latest Publications
- 26 March 2024
- Published by Springer Nature
- p. 178-189
ELECTRA: A Comprehensive Ecosystem for Electric Vehicles and Intelligent Transportation Using YOLO
- 26 March 2024
- Published by Springer Nature
- p. 178-189
ELECTRA: A Comprehensive Ecosystem for Electric Vehicles and Intelligent Transportation Using YOLO
- Open Access
- 3 August 2023
- Published by Association for Computing Machinery (ACM)
- p. 467-473
U-Net and YOLO: AIML Models for Lane and Object Detection in Real-Time
In recent years, significant research and funding have been directed towards the creation of self-driving cars. Real-time object and lane detection on roadways is a crucial aspect of autonomous driving systems. In this research, we offer a software-based self-driving vehicle system that detects objects and lanes using computer vision techniques. Our system can detect variety of things, from vehicles to lanes, and is built to achieve results with as little latency as possible. Currently in detection algorithms there has a significant amount theory suggesting the use of CNNs (Convolutional Neural Network) which learns and generates arrays such as feature maps. Multiple layers of these layers generate parameters that help the next learning layer and are therefore recognized for their exceptional performance in recognition and detection tasks. Using actual driving circumstances, we evaluate our approach and discuss its challenges and limitations. We also take into account prospective strategies for improving the system's performance and functionality in the future. Here we have followed a U-Net model architecture which calls features from earlier Convolutional layers unlike the normal neural network architecture and therefore is considered to be one of the best models for segmentation tasks. Object Detection has recently gained a lot of attention as Drones and other devices which need camera feedback has increased, and in this there has already been a lot of research conducted such Mask-RCNN, Faster-RCNN, etc. but out of all the Networks, the benchmark which YOLO has set is record breaking. Multiple little variations of YOLO are available in market because of it relatively higher speed and accuracy then others. Our research has implemented these models and can help develop secure and efficient transport networks, and it has substantial implications for the development of autonomous driving technologies. Results show that U-Net has competed with a good rmse score. YOLO model requires relatively low complications and has shown Higher accuracy in the the field of object detection till date. We have tried and tested with CULane dataset for Lane Detection as the data provides lane markings as annotation while Custom Dataset for our object detection that mainly focuses on identification of classes related to traffic ecosystem model.
- 3 August 2023
- Published by Association for Computing Machinery (ACM)
- p. 467-473
Open AccessU-Net and YOLO: AIML Models for Lane and Object Detection in Real-Time
In recent years, significant research and funding have been directed towards the creation of self-driving cars. Real-time object and lane detection on roadways is a crucial aspect of autonomous driving systems. In this research, we offer a software-based self-driving vehicle system that detects objects and lanes using computer vision techniques. Our system can detect variety of things, from vehicles to lanes, and is built to achieve results with as little latency as possible. Currently in detection algorithms there has a significant amount theory suggesting the use of CNNs (Convolutional Neural Network) which learns and generates arrays such as feature maps. Multiple layers of these layers generate parameters that help the next learning layer and are therefore recognized for their exceptional performance in recognition and detection tasks. Using actual driving circumstances, we evaluate our approach and discuss its challenges and limitations. We also take into account prospective strategies for improving the system's performance and functionality in the future. Here we have followed a U-Net model architecture which calls features from earlier Convolutional layers unlike the normal neural network architecture and therefore is considered to be one of the best models for segmentation tasks. Object Detection has recently gained a lot of attention as Drones and other devices which need camera feedback has increased, and in this there has already been a lot of research conducted such Mask-RCNN, Faster-RCNN, etc. but out of all the Networks, the benchmark which YOLO has set is record breaking. Multiple little variations of YOLO are available in market because of it relatively higher speed and accuracy then others. Our research has implemented these models and can help develop secure and efficient transport networks, and it has substantial implications for the development of autonomous driving technologies. Results show that U-Net has competed with a good rmse score. YOLO model requires relatively low complications and has shown Higher accuracy in the the field of object detection till date. We have tried and tested with CULane dataset for Lane Detection as the data provides lane markings as annotation while Custom Dataset for our object detection that mainly focuses on identification of classes related to traffic ecosystem model.