Project Idea| CamCaliber

Project Title: CamCaliber Introduction: Camera calibration is an important task in computer vision applications. In a video surveillance system with a single static camera, object tracking results of moving objects can be effectively used for camera calibration. Although Cam-calibration from object tracking has been studied for years, it is still facing many challenges. The performance of cam-calibration is highly dependent on the accuracy of extracted head and foot locations, which is related to the robustness of the segmentation and tracking approach. Therefore, we have tried to use a robust object segmentation and tracking system to achieve accurate head/foot localization. They are extracted based on the tracking result and segmented foreground blob of each object. Conceptual framework: We propose a method to design a system in such a way that it can optimize tracking results using deep learning techniques i.e. convolutional neural network (CNN) which is a fast object detection method. In order to achieve our project, we use a sequence of positive images and negative images of an object to be detected and find the result efficiently. We implement this using TensorFlow library in Python language. This system can be implemented using sensors to work upon tracking objects remotely, for example, self-driving cars, moreover, thief catching, Driver assistance etc. With this, it is done efficiently and is shown in snapshots below. Objective:  The Objective is to design a robust and efficient system to solve object tracking issues in the field of video surveillance as well as real-time video analysis. Present issues: The recent studies show that in video surveillance there are challenging issues for tracking objects like camera rotation, false localization etc. Let's say in a Live match, an object is to be tracked i.e. a ball so, during the live match, a camera holder wouldn't be able to track when hit strongly. Hence, the camera is made in such a way that it will capture itself and show the resulting ball by ball. The real-time cost will be less and efficient. Snapshots: Tools Used:

• Sensors (Send data privately to the Cloud storage)
• HQ Camera
• Python 3.4+
• OpenCV (Open Source Computer Vision Library)

For this, we have used deep learning techniques to model this using CNN(Convolutional Neural Network) with TensorFlow library. Future Enhancements: In future, although we can implement in many video surveillance application, it can also be implemented in various platforms of detecting and tracking objects like Car racing, Motor Bike racing, Live matches like Cricket, Football to assist the players, viewers during the matches. In addition, we can also implement this anywhere to track any kind of objects and using sensors, we can get information privately from the cloud and make the result remotely. GitHub Code Repository Link: https://github.com/afzal442/CamCaliber Team members:

• Afzal Ansari
• Saan

Note: This project idea is contributed for ProGeek Cup 2.0- A project competition by GeeksforGeeks.