.png){kind=small}
The theft of catalytic converters has become an increasingly prevalent issue, as reported by Fight Rising Catalytic Converter Theft | NADA. Instances of catalytic converter theft have surged from 16,660 in 2020 to a staggering 64,701 in 2022, resulting in substantial financial burdens for vehicle owners who face replacement costs ranging from $1,000 to $3,500.
Utilizing a computer vision model trained on a dataset of 2 billion images sourced from across the Internet, EyesOnIt is equipped to identify incidents such as catalytic converter theft. When integrated into existing security systems, EyesOnIt serves as a powerful tool alongside other deterrents to prevent perpetrators from causing damage to vehicles.
The image below illustrates EyesOnIt's capability to accurately detect catalytic converter thieves through intuitive object descriptions.
To understand these results, let's review four concepts that are central to EyesOnIt detection:
Text descriptions – detection is based on English object descriptions
Multiple object descriptions – two or more object descriptions are needed for accurate results
Full image comparison – EyesOnIt compares each object description to an entire image or to each selected tile.
Confidence levels – EyesOnIt reports the match between an object description and an image as a confidence level with 0% (no match) and 100% (strong match). The confidence levels for all descriptions will add up to 100% for a single image. When multiple tiles are selected, EyesOnIt reports the maximum confidence level for each object description across all tiles.
For a detailed explanation of these concepts, refer to our previous post: "Deep-Dive Into the EyesOnIt Individual Image Configuration User Interface."
Combining these key concepts enables EyesOnIt to provide precise detection of catalytic converter theft using the object descriptions shown above:
Empty parking space – High confidence indicates the absence of a car, eliminating the need for an alert.
Car – high confidence for this description means that a car is present. Since other descriptions also include the word “car”, EyesOnIt will only give “car” high confidence if it is more accurate than the other descriptions. In other words, “car” will have high confidence if there are no people near or under the car. No alert should be sent if EyesOnIt just sees a car.
Person under car – this object description is quite specific. This description will have high confidence if EyesOnIt sees a car with a person under it. We definitely want an alert in this case. If there is a person near the car, or a car with no person, the confidence level for this description will be low.
Person near car – this object description is also pretty specific. The confidence level here will be high if a person is near the car but not under the car. Each user or installer would decide whether they want an alert for this condition given their specific circumstances.
As a side note, the video alerting configuration UI allows users to set alerting criteria including confidence levels and time thresholds. Installers can implement more advanced alerting logic with the EyesOnIt REST API or with help from EyesOnIt developer support. EyesOnIt can also be configured to automatically select the right tile or tiles for description comparison using motion detection.
Based on a high-confidence detection that a person is under or loitering near a car, EyesOnIt can send an SMS alert or trigger an action in another system by calling a REST API endpoint.
Catalytic converter theft often occurs in low-light scenarios. The subject is a little tougher to see in the images below, but these pictures show that EyesOnIt is just as successful in these harder cases.
These are just a few examples of how the EyesOnIt pre-trained model can be configured to detect catalytic converter theft. The same approach can be used in much larger car lots owned by rental car companies or auto dealerships. EyesOnIt can also be used in related scenarios to detect conditions that will improve security, safety, and efficiency.
Contact us today to discuss how EyesOnIt can help with your use case.