Sam Armani, PHD

A new spatial optimization algorithm is proposed to assure over a certain percentage of emergency calls will be responded on time while the operation cost is minimized. The best locations for posting a vehicle as well as the number of vehicles required to cover the area during each hour of the day provided as two key information for deployment planning. A combination of hardware and software speeding up solutions such as developing fast search algorithms and parallel processing techniques used to deploy the prototype system in the company production system.

A pilot project to assess the viability of using machine vision techniques to automate report writing of the company's field operations. Accordingly, Bags of Visual Words, Convolutional Neural Networks, and Mask-RCNN methods performance were tested in real-time detection. ConvNet architecture, layers, and hyperparameters were briefly studied using Keras with TensorFlow backend. Three Deep Learning boxes consist of a different hardware setup were used during the training stage with near a million sample images to identify cost-effective hardware investments.

Developed a Python tool to analyze the accuracy of a computed drive-time ring based on available options using real-world historical traffic data. The estimations methods were compared against nation-wide major cites historical data in various time-context such as hour of day and day of week to find out the most suitable method for the company's business. In addition, the selected drive-time estimation method was calibrated using 20000000+ ambulance trajectory data collected during companies emergency transportation services to further improve the accuracy.

A framework is proposed in order to automate and improve traffic data quality control for IDOT. The framework consist of data verification and data validation procedures. In data verification process the main idea is to check with data consistency between data and original copy of the data. In data validation process the main idea is to make sure that the verified data are valid data. Traffic sensor malfunction, inclement weather condition, and road events such as work zones as well as traffic accidents are sample reasons that may lead traffic engineers to ignore the collected data.

Credit: Shu Yang (PhD Student) for his implementations. This project developed a database specifically designed and optimized for the purpose of freeway performance measurement and implemented a data quality assurance procedure to examine the quality of the data collected. After a comprehensive literature review, the instantaneous travel time estimation model was selected and implemented in a stand-alone MATLAB-based system and a Microsoft Excel VBA-based tool. Several case studies were conducted to test the applicability of the system.

A web-based GIS application for analyzing traffic counts for ADA, Riyadh, Saudi Arabia is developed. The underlying database is optimized to support the application queries with fast responses. Various charts including combo charts, treemaps, and calendar charts is used to visualize the traffic counts. Several features including time-based traffic counts, location-based traffic counts, and spatiotemporal summaries are implemented to support next traffic engineering analysis. A live demo of the application is available in adatraffic.com.

A new algorithm is proposed to improve V2V communications in the connected vehicle environment. Accordingly, Michigan CV Test bed is modeled in PTV VISSIM, and the simulation results pushed into Microsoft SQL server. Then, a Python program integrated with ArcGIS functionalities is implemented to evaluate our proposed method efficiency against previous methods.

A customized Genetic Algorithm is implemented in Matlab environment to detect optimal threshold values for our proposed classifier. In this accord, all the possible threshold values in the poposed model tested against a dataset with labeled classes. The objective function is set to maximize the model sensitivity, while the number of false positives are minimum.

Credit: Professor Kianfar for his advice. Several Unsupervised Machine Learning methods has been investigated to evaluate the most accurate approach based on different input variables for automatic Traffic Fundamental Diagrams generation. The proposed method also automate extracting key traffic stream characteristics including PQF, QDF, and critical occupancy.