Research Projects

With the advancements in the field of autonomous vehicle technology including drones, there is a requirement for highly efficient and reliable decision-making systems that are capable of ensuring safety. This project leverages Generative Adversarial Networks in order to enhance the ability of self-driving vehicle or drone to process visual inputs and make real-time decisions. First, we create a synthetic image by fusing inputs taken from multiple cameras positioned at various angles on the vehicle (movable as well as immovable). This image must provide clarity about the vehicle’s next move, hence improving the decision-making process. Our primary goal is to reduce the latency in producing the image and optimize the overall performance. The GAN based framework here will combine the inputs from different cameras to generate an image with maximum information and accuracy. By incorporating technologies like Convolutional Neural Networks and real-time data processors, we ensure that the process has minimal delay. We also explore the potential use of hardware techniques such as GPUs to further improve system performance. Self-driving vehicle or drone can equipped with advanced technologies like radar, HDR cameras and ultrasonic sensors to perceive their environment correctly and make decisions without human intervention. These systems must collect and analyze data in real-time to ensure safe interaction with conditions, other vehicles and pedestrians. One key aspect of autonomous driving is image processing, where inputs from multiple cameras at different angles are captured and processed in real time. The traditional methods of processing multiple inputs often lead to latency in decision-making, which may compromise the safety of drivers. In order to address this challenge, GANs have emerged as a solution. GANs consist of two neural networks- the generator and the discriminator, working together to produce realistic images. Through an adversarial process, GANs can generate high-quality images that combine multiple inputs into a single representation. Here, we will use them to merge input images from various cameras, providing the vehicle with a clear view for better decision-making. The proposed solution addresses the issue of latency in autonomous driving systems, which currently struggle to process data quickly enough to make efficient decisions. By integrating deep learning techniques and optimizing the system architecture, our goal is to develop a scalable solution to be implemented in real-world vehicles.