🖥️Data Processing

✨ Photo Quality

The first step on our data processing is to filter out blurred and unusable pictures. Based on information collected from the different phone sensors and the sharpness of the photo, we calculate if the photo can be passed to our neural network or if the photo is not eligible.

Blurred/unusable image

🔍 Photo Verification

The process of photo verification utilizes advanced image processing algorithms to filter out invalid images and detect potential cheaters.

Additionally, the system may randomly select images for manual verification, where members of the community can participate and earn rewards in the form of $CITY tokens for their efforts. To prevent dishonest contributors, the manual verification process employs "cross-validation" techniques to ensure that a diverse group of users are evaluating each image, and any participants found to be providing fraudulent feedback will be penalized.

Invalid photos may include:

• A photo of a person's face

• A photo of a private room or personal property

• A photo of a non-urban area such as a forest or a lake

• A photo taken from a video

It's important to remember that City Sight's data collection process focuses on collecting images of public locations and objects, such as public roads, buildings, street signs, and landmarks.

All photos passed this stage are eligible for rewards in the form of $CITY tokens. More details can be found on the #Contribute2Earn section

🧠 Object Detection

On the City Sight project, we use advanced machine learning techniques to analyze the data collected from our users. By training our neural network on a large dataset of images and labels, we are able to accurately detect a wide range of objects in each photograph. This includes people, vehicles, ads, cyclists, logos, buildings, signs, and roads. The object detection capabilities of our neural network are highly sophisticated, allowing us to detect even subtle features and patterns that may be difficult for humans to discern.

Once the objects have been detected, we assign a confidence score to each one based on the accuracy of the detection. This score is used to filter out any false positives or low-confidence detections, ensuring that the data we provide to our clients is of the highest quality.

Example of people detection

🔢Data Modeling

With the data collected from our object detection process, we are able to generate detailed models of how people and objects move and interact in different environments. By inputting this data into our models, we are able to predict and analyze patterns of behavior and changes in data on a large scale.

This allows us to provide valuable insights to our clients, including businesses and organizations looking to understand consumer behavior and make informed decisions about location and marketing strategies. It also allows us to provide tailored solutions to governments and other organizations responsible for maintaining and updating road maps and infrastructure, helping them to keep their maps accurate and up-to-date in an increasingly connected world.

People counting from location within 10 meter distance