Project 1: Autonomous Vehicles' Traffic Sign Detection using TensorFlow and OpenCV
Embedded Files
Detected traffic sign
As reported in the findings of an eight-year study conducted by the National Highway Traffic Safety Administration (NHTSA), there were on average 1,578 fatalities each year resulting from two-vehicle traffic crashes at intersections controlled by traffic signals. Approximately 51% of those fatal crashes were caused by drivers who ran red lights. Approximately 29% were caused by drivers who failed to yield the right-of-way at traffic signals." These lines taken from the NHTSA study shows the grim number of deaths that occur each year due to the misinterpretation of traffic signs ultimately caused due to human error. Although not completely, we believe that the deployment of machine learning-based models in cars to assist drives will help reduce the number of accidents occurring due to human error substantially. In this project, we present a unique thin yet deep convolutional neural network architecture for traffic sign identification that is both energy-efficient and robust. Each convolutional layer in the proposed design has fewer than 50 features, allowing us to train our convolutional neural network without necessarily using a graphics processing unit.
Project 2: Autonomous Vehicles' braking System
In this project that I have created, The model is able to successfully read to distances between 4 to 100 cm and modify the speed of our motor through voltage pulsing within a span of 140 ms on average. When the distance is beyond the declaration threshold, the motor is pushed to peak voltage which increases its speed and when it falls below the deceleration threshold the voltage is reduced and the speed of the motor decreases. Our 2-month long project and research have shown that smart systems can be designed and implemented with relative ease for most mass-produced vehicles. This is in contrast to popular trends in the vehicle market which reserve safety and convince features such as the dynamic braking system for expensive high end autonomous or semi-autonomous vehicles. This study also analysed the effectiveness of a mono-sensor based system, and found that while these systems may excel in certain scenarios in performing one task, it is best to rely on multiple sensors to accomplish the same task as it covers a wider array of data and can lead to better effectiveness of the task carried. We believe future work could seek to integrate and test a combination of sensors and study the results.
Project 3: SenseNet - Posture Correction and Assistance System for Handicapped and Visually Impaired
Page updated
Google Sites
Report abuse