RecallRoute

No routes are forgotten.

Comment

Project / idea name

RecallRoute: AI-Powered Search Intelligence for Missing Alzheimer’s Patients

  • RecallRoute is an AI-powered emergency search system designed to help locate missing Alzheimer’s patients as quickly as possible.
  • The system combines the patient’s past routines, familiar locations, approximate mobile-network location data, witness reports, maps, and search results into one continuously updated model.
  • Instead of treating the case as a single point on a map, RecallRoute estimates how the patient’s probable location may change over time and identifies where search teams should look next.

Track

AI

Problem statement: What real-world problem are you addressing?

  • Alzheimer’s patients can become disoriented and may leave home without understanding where they are going or how to return.
  • However, their movement is not always completely random. They may continue along familiar roads, visit places they recognize, or follow routines from their past.
  • During a missing-person case, information usually arrives from many different sources, including family members, witnesses, mobile networks, maps, and search teams.
  • These sources may be incomplete, delayed, or contradictory, making it difficult to determine which areas deserve immediate attention.
  • A conventional map can show possible locations, but it does not adequately represent how a person may move through connected roads and pathways over time.
  • The central challenge is therefore to continuously estimate where the patient is most likely to be, how that probability may change, and where search efforts should be directed next.
  • In such situations, every minute matters. A system that can organize evidence and reduce the search area faster could significantly improve the chances of finding the person safely.

Proposed solution: What is your idea and how would it work?

  • When a patient is reported missing, RecallRoute creates an initial probability map using their last known location, known routines, familiar places, historical movement patterns, and surrounding geography.
  • Authorized mobile-network information can provide an approximate location of the patient’s phone, which is treated as evidence rather than an exact position. Based upon the timestamps, the direction of movement can be approximated.
  • Family members and witnesses can submit sightings with the time, location, direction of movement, and other useful observations.
  • AI analyzes this information and identifies likely movement patterns based on the patient’s personal history.
  • The city is represented as a network of actual roads and pedestrian paths, allowing the system to model realistic movement instead of spreading probability across inaccessible areas.
  • As new information arrives, the probability map is updated continuously. Conflicting information increases uncertainty instead of forcing the system to choose an unreliable answer.
  • Search results are also included in the model. If a team searches an area without finding the patient, the system reduces the probability of that area according to how effectively it was searched and shifts attention toward remaining possibilities.
  • The system maintains a chronological record of events so that investigators can see what was known, when it became known, and how each piece of evidence changed the search.
  • The final interface provides search teams with probable locations, important movement corridors, confidence levels, and recommended areas to investigate next.

Technology component: How does technology, science, or engineering contribute to the solution?

  • Advection-Diffusion Fokker-Planck Model on a Metric Graph: RecallRoute models the road and pedestrian network as a metric graph, where streets are connected segments with real physical lengths. A Fokker-Planck equation is then used to describe how the probability of the patient being at different points on this network changes over time. $$-\frac{\partial}{\partial s}(v_e\rho_e) + D_e(t)\frac{\partial^2\rho_e}{\partial s^2}$$

Here, the advection term represents directional movement toward likely destinations, while the diffusion term represents uncertainty and unpredictable movement. This allows probability to move through actual walkable routes rather than spreading freely across a normal 2D map.

  • Personalized Behavioral AI: The system learns from information such as familiar destinations, previous routes, time of day, and repeated movement patterns. These patterns create a personalized probability model for the individual instead of assuming that every missing person behaves in the same way.

  • Subjective Logic for Evidence Fusion: Witness reports, family information, and mobile-network observations can have different levels of reliability. Subjective Logic represents each source using belief, disbelief, and uncertainty, allowing RecallRoute to handle conflicting evidence without creating false confidence.

  • Continuous Updating from Search Results: A failed search is also useful information. RecallRoute models the search area and the effectiveness of the search before reducing its probability. This prevents the system from simply deleting an area from the map and allows probability to shift toward locations that remain unexplored.

  • Information Gain for Search Decisions: RecallRoute estimates which available search action or new piece of information could reduce uncertainty the most. This helps teams decide not only where the patient may be, but also what they should investigate next.

  • AI-Based Search Interface: The technical models are presented through a simple map-based interface showing probability zones, likely movement corridors, evidence confidence, recent updates, and recommended search priorities.

Target users or beneficiaries: Who would benefit?

  • Police departments and emergency response teams could use RecallRoute to organize large amounts of information during missing-person investigations.
  • Search-and-rescue teams could use the probability map to prioritize areas and reduce unnecessary coverage.
  • Families and caregivers could benefit from faster coordination and better organization of information during an extremely stressful situation.
  • Hospitals, elder-care organizations, and community support groups could potentially use the system as part of emergency response procedures.
  • The underlying approach could later be adapted for other vulnerable missing-person cases while keeping the primary focus on Alzheimer’s patients.

Potential impact: What could improve if the idea were implemented?

  • RecallRoute could reduce the time required to organize scattered information and turn it into an actionable search plan.
  • Personalized behavioral prediction could help search teams focus on routes and destinations that are more relevant to the individual.
  • The metric-graph movement model could prevent search resources from being wasted on physically inaccessible areas.
  • Uncertainty-aware evidence fusion could reduce the risk of overreacting to a single unreliable sighting.
  • Learning from failed searches could prevent teams from repeatedly concentrating on the same locations.
  • Information Gain could help teams prioritize actions that are most likely to improve the search.
  • Most importantly, the system is designed around a simple objective: reduce the time between a person going missing and responders reaching the places where they are most likely to be.

Feasibility / implementation plan: How could this idea realistically be developed?

  • Stage 1: A digital map of a selected city can be converted into a metric graph containing roads, pedestrian paths, intersections, and approximate travel distances.
  • Stage 2: Synthetic missing-person scenarios can be created to train and test the behavioral AI without requiring sensitive personal information.
  • Stage 3: The Fokker-Planck movement model can be implemented and tested against simulated walking patterns to determine whether probability remains concentrated around realistic routes.
  • Stage 4: Witness reports, family information, and simulated mobile-network observations can be introduced with different reliability levels to test the Subjective Logic evidence-fusion system.
  • Stage 5: Simulated search teams can be used to test how negative search results change the probability distribution and whether Information Gain improves search decisions.
  • Stage 6: These models can be integrated into a live map interface that updates as new information is entered.
  • Stage 7: With appropriate partnerships and legal authorization, the prototype could eventually be connected to real emergency-response systems and authorized mobile-network data.
  • The initial prototype can therefore be developed entirely with maps and simulated data, making the concept technically testable before any access to sensitive real-world data is required.
  • Performance can be evaluated using measures such as localization error, probability assigned to the true route, reduction in search area, reduction in uncertainty, and time taken to identify the correct search corridor.

Challenges or limitations: What obstacles would need to be addressed?

  • Mobile-network data may only provide an approximate location and may become less useful when the phone is switched off or disconnected.
  • A patient may behave unpredictably, meaning behavioral AI can provide probabilities but cannot guarantee a destination.
  • Witnesses may provide incorrect locations, times, or directions, especially during stressful situations.
  • Map data may be incomplete or may not accurately represent temporary barriers, private areas, or inaccessible paths.
  • A failed search does not always mean that the patient was absent from the searched area, so search effectiveness must be estimated carefully.
  • The system would require strict privacy controls because location and personal behavioral information are highly sensitive.
  • Access to real mobile-network data would require cooperation with authorized telecom and emergency-response organizations.
  • AI recommendations should support trained responders rather than replace human judgment.
  • RecallRoute would not guarantee a successful rescue. Its purpose is to make the search process faster, more organized, and more evidence-driven.

AI Usage Disclosure

The ideation, problem identification, research direction, feature selection, and overall concept development were done by me. AI was used to assist with organizing and clearly writing the project description and presenting the ideas in a structured manner. The technical concepts and proposed features were selected and defined as part of my own project ideation.

Share this project:

Updates

Submission history