Join us to participate in the upcoming 2019 City:One Challenge. 

Predict Pedestrian Behavior

Developing solutions through AI to predict/anticipate pedestrian behavior in the urban environment resulting in improved pedestrian safety.

Photo of Intvo LLC
0 2

Written by

Autonomous vehicles have the potential to create a world with fewer limitations, greater accessibility, and safer roads. As the automotive industry develops autonomous vehicles, one of the major hurdles they face is ensuring the safety of road users. Greater distractions, especially from digital technologies, are responsible for a recent rise in pedestrian deaths. Here, at Intvo, we are developing artificial intelligence that predicts pedestrian behaviors. Our product goes beyond processing physical movement vectors in space. It is designed with human factors in mind, allowing it to account for body language and human reflexes to decode both intended behaviors and slips that occur as the result of distracted pedestrians.

Intvo’s software is innovative, unique, and most importantly, its core purpose is improving safety and saving lives. Efficient and safe navigation in traffic requires cognitive skills, risk prediction ability, and ethical decision-making. Humans already possess these skills and abilities. We are developing software that will resemble the abilities of the human brain and even outperform them.

Our Vision is to strengthen the safety, confidence, and trust for self-driving vehicles with the help of AI. We are striving to ensure a safe road environment and ease the lives of all individuals. Our smart vision software pushes the boundaries of autonomous technology in the existing market and enhances the robustness of the autonomous vehicle’s vision.

Describe who will use your solution (1,000 characters)

Our innovation integrates perspectives of moving vehicles and/or connected vehicles in an infrastructure to generate a much better understanding of human behavior than either perspective alone

Describe your solution's stage of development

  • Pilot - you have implemented your solution in a real-world scenario

Insights from previous testing (500 characters)

Leading into customer engagement, Intvo has confirmed pilot partnerships with the University of Michigan Transportation Research Institute, and the Toledo Area Transit Authority to gather data and test driver alert systems that will lead to potential vehicle interaction in the future.

Tell us about your team or organization (500 characters)

Intvo is a startup based in Ann Arbor, MI with an enthusiastic team comprised of the following key members: Assam Alzookery – Founder & CEO Sharena Rice – Senior Behavioral Scientist Brian Korzynski – Senior Data Scientist Anthony Bozzini– Head of Business Development Joaquin Nuno-Whelan - Strategy Advisor

Size of your team or organization

  • 2-10

Funding Request

  • $100,000

Rough Budget (500 characters)

Deploying this innovation will require a rough budget of $85K which will be used for hardware, installation, testing and staff cost.

Describe how you would pilot your idea (1000 characters)

We plan to achieve this project’s goal in two objectives, summarized below.: Objective 1: We will spend approximately 150 hours collecting data which will be used for further testing in our lab. Intvo will use the collected data to retrain our behavioral model. The software will then be tested at MCity for 85% or greater accuracy before implementing it. Objective 2: after 4 months, Intvo will deploy its pedestrian prediction software and hardware. Intvo software will perform pedestrian path prediction that will be sent to the Roadside Units (RSU). The RSU will construct and broadcast the pedestrian safety message (PSM) to any connected vehicle and inform/warn the driver appropriately. Intvo’s team will test the software’s accuracy on the infrastructure. The software will be constantly monitored by the Intvo team and on-site checks will be conducted once a week.

Describe how you would measure the success of your pilot (1000 characters)

The measured system false-positive performance baseline is 85% accuracy. We will decrease the false positives to the level agreed to by the team with a target of 90% accuracy. We will check with the following parameters in mind: Frequency - with each time the simulation model is run. Resolution - in use cases including day, night, weather, etc. Accuracy - using false positive criteria that partner and Intvo agree upon. Tools - recorded video and shared classification sessions with partner and customer with access to drivers for interviews. With the data we are collecting, we will also define, enhance, and measure path prediction performance over time. We will start with baseline data including intersection cameras. We will demonstrate the improvement with a re-trained model. Our goal for metrics is to demonstrate our ability to improve safety with pedestrian behavior signal accuracy from a baseline of X% to XX% over a 6-month span.

Sustainability Plan (500 characters)

Yes, Our goal is to continue improving our models to help improve pedestrian safety


Join the conversation: