GA8ED

Real-Time Incident Detection for Neighborhood Safety

This student team will work to build an action and incident classification system that utilizes live camera feeds to detect safety issues in residential neighborhoods. The system should utilize multiple cameras to cover large areas, an edge embedded and cloud based inferencing sub-system that raises alerts when it detects safety concerns. The raised alerts should contain a description of the safety concern. The project should be built leveraging already existing technologies as much as possible and clearly identify areas that need to be developed. The system should have no more than a 10 second delay from the first occurrence of a safety issue to the notification of that safety issue. The following safety concerns would be targeted (listed in increasing order of difficulty): speeding detection, bear sighting, car prowler detection. In the case of animal detection, alerts should not be raised for pets / deer. The false positive (FP) and false negative (FN) rates for each of the safety issues should be under 20% and 5% respectively (initial goal at project completion). Students with work with mentors to synchronize on requirements, propose a design, build a prototype system, produce a final report of learnings, provide an end-to-end demonstration and present it at the capstone fair. Students should be able to validate and test the system both in lab settings and in real world settings (location will be provided).