Mobile Damage Detection

Two different detectors were implemented to address screen and back panel defects. The screen defect detector was built using techniques like adaptive thresholding, shape detection, Hough Transform, and morphological operations. Meanwhile, the back panel detector leveraged technologies such as Pytesseract and Mask RCNN. To mitigate fraud, several checkpoints were put in place, including skin detection, complete phone extraction, excess brightness detection, and glare detection

The story

The project involved the automation of physical damage detection in mobile phones. The primary goal was to develop a model capable of classifying and localizing different defects on the mobile phone’s screen and back panel, such as broken screens, spots, and dents. This system would also estimate the extent of damage, categorizing it as minor or major, which would have a direct impact on the mobile’s price.

The challenge

The primary challenge was to create a reliable and efficient system for detecting physical damage on mobile phones. This involved developing models for both screen and back panel defect detection. Furthermore, the system needed to estimate the extent of damage, classifying it as minor or major, which would impact the mobile’s price. Making this system work in real-time and under various lighting conditions added complexity to the project.

Additionally, ensuring the integrity of the system and mitigating potential fraud was a crucial challenge. This necessitated implementing various fraud detection mechanisms, such as skin detection, complete phone extraction, excess brightness detection, and glare detection.

The solution

To address these challenges, a combination of technologies and tools were employed. For the screen defect detection, adaptive thresholding, shape detection, Hough Transform, and morphological operations were used. For back panel defect detection, Pytesseract and Mask RCNN were utilized. The use of Python, Keras, Tensorflow, SSD Mobilenet CNN, AWS, S3 bucket, and scikit-learn played a vital role in building the system.

A mobile app was developed to make the system accessible to users, and Flask was used to create a user-friendly interface. OpenCV was crucial in handling image processing tasks. The comprehensive solution ensured the accurate detection and classification of physical damage on mobile phones, offering a valuable tool for mobile sellers and buyers.

The outcome

The project resulted in a successful implementation of a physical damage detection system for mobile phones. This system provided accurate detection and localization of defects on both the screen and back panel. It also estimated the extent of damage, categorizing it as minor or major, which had a direct impact on the mobile’s price.

The mobile app created for this purpose was robust and capable of running the model in real-time and under various lighting conditions, making it highly practical for use in different environments.

Moreover, the fraud detection mechanisms put in place, including skin detection, complete phone extraction, excess brightness detection, and glare detection, added an extra layer of security and integrity to the system.

In summary, the project successfully automated the process of detecting physical damage in mobile phones, providing a reliable and efficient tool for assessing the condition and value of mobile devices.

Planning to outsource software development services?
Contact sales, to start a project, now.