Building a Vehicle Recognition and Inspection System Using Computer Vision

Develop a vehicle recognition and inspection system using captured pictures and videos of cars. In this 8-weeks challenge, you will join a collaborative team of 50 AI engineers from all around the world.

The problem

The real-world problem that is being addressed in this challenge is car theft and fraudulent activities aided by changing vehicles.

Public safety, traffic operations, transportation planning, and other essential choices made by road safety and transportation organizations rely significantly on traffic data collected. Magnetic loops, pneumatic road tubes, and piezoelectric sensors are examples of traditional traffic data-gathering technology that have been on the market since the 1900s. However, modern video analytics tools based on computer vision architectures now allow video surveillance systems to replace older systems as a better, more dependable, and less expensive alternative.

The solution is a computer vision-based project. The goal is to capture a live image of a vehicle with both time and geo stamp embedded in the file, blocking PnP manipulation in line with the use cases detailed under the project goals.

The project goals

The goal of this AI challenge is to recognize a vehicle, validate that it’s the same as the previously captured vehicle, and identify scratches, dents, and damages to parts of the car with the severity of each damage type via analyzing pictures and videos of the car captured.

• Pre-loss inspection

The process is an inspection of the vehicle capturing and storing the vehicle’s identity, as mentioned above, blocking any form of user manipulation. Scratches and dents should also be captured during this phase.

• Post-loss inspection

During the post-loss inspection (after damage), the user is expected to conduct the same inspection process and ensure the damaged parts are captured. All manipulations also are to be blocked just as the pre-loss.

• The models to be built will compare the two vehicles for validation and match i.e the system will compare if the post-loss vehicle is the exact same vehicle as the one the pre-loss inspection.
• The system will identify the damaged part(s), and type else will indicate the absence of damage.
• Another layer is to identify the extent/severity level of the damage.
• The final step would be to give a cost estimation of damaged parts.

The solution should be models/APIs testable and can be consumed/deployed into any application.