(searched for: doi:10.7546/ijba.2021.25.1.000611)
Published: 29 June 2022
International Journal of Advanced Research in Science, Communication and Technology pp 433-438; https://doi.org/10.48175/ijarsct-5368
In this paper, we present a scheme for traffic analysis using Image Processing to alert traffic control. In this, the vehicles are not being detected by sensors as we are detecting by images with the use of python language we are going to implement it in our project. Once image is captured from digital media, it is fed into image processing after that it detects vehicles from image using open cv libraries, after that at the end vehicles are detected on basis on vehicle count, and time will be set as per so reduce the road traffic congestion. This system contains the solution to three problems of traffic system. First one being the pre-defined set of timings set for each traffic signal despite the density circumstances. For this we have changed the signal timings. The working would be as follows, in a traffic junction of four lanes the density is measured on each lane at distance of 50 meters through the Image Processing. After that count the vehicle and turn on green light for time period deepening on vehicle count ratio. this is done so that the lane having highest density is allowed to clear the traffic first, the other lanes will be given green signal after this in a circular pattern. If in cases where the density is greater, the signal timing is increased seconds
Published: 24 May 2022
Conference: 2022 1st International Conference on AI in Cybersecurity (ICAIC), 2022-5-24 - 2022-5-26, Victoria, United States
The stock market is one of the most important topics of today’s economy due to its fluctuating nature and far-reaching impact. Despite the difficulty of mathematical modeling of time-series of stock prices, deep learning and black-box modeling have been reported to perform well in literature. In this work, we have experimented with different recurrent neural network architectures to predict stock prices for elite Nasdaq companies. We have improved the baseline models by reducing the number of parameters and using bidirectional processing with attention. We show that using a Gaussian noise vector can regularize the model to improve the robustness without using any dropout. Our proposed model achieves the lowest mean square error for AMZN (0.12), AAPL (0.15), NFLX (0.03), GOOG (0.08), GOOGL (0.06), CSCO (0.01), COST (0.06), FB (0.03), 8 out of 10 companies. Our work aims to show that with better architecture design, RNN variants can outperform baseline models despite having a fraction of the parameters.