Abstract

Be it in any industry and innovation, AI can work many wonders. One such marvel is the detection and measurement of the height of an individual from an image. In Physical Efficiency tests, manual height scale measurement along with computer vision technology can enhance the overall credibility of the procedure

Problem

Determining the exact height of a candidate in the process of PET has a lot of complications where prominent ones are inaccuracy, corruption, or no proof of data that has been collected. With this, the process of height calculation has become a tricky exercise.

Solution

Deployment of AI: We aim to measure the height of candidates from a photograph taken during the Physical Measurement Tests of Civil Force examinations run across the country. To perform this, we chose an economical way by using the right equipment with automatic response. A measuring scale is physically put beside the candidate during the process. A photograph is captured from which the exact height of the candidate is calculated through AI, referring to the adjacently placed measuring scale.

An automated approach through Computer vision is employed in this case to interpret the visual data (images/videos). Optimal distance is maintained between the camera and the candidate to capture the image correctly and with exceptional quality. Here, Convolutional neural networks divide every image that needs to be tagged into smaller sections.

Key Functions:

1. Data Collection: Candidate data in the form of images is retrieved from the examination center. Relevant image datasets are collected for tagging the data points in image graphs. The image data is then deployed into a selected computer vision system.

2. Annotation: Through this process, the collected candidate image data is labeled with the identified key points and landmarks. The images are converted into point graphs, from which the points of the image are retrieved to attain the exact height point of the candidate from the image.

This process recognizes, counts, and tracks the images. All the retrieved points from the image are stored in an XML file of matrices.

3. Training: The system is trained using transfer learning to conveniently identify from the collected datasets. A relevant model is built in such a way that the obtained data points from the pool are essentially computed based on the Xmin, Xmax, Ymin, and Ymax operations which ultimately determine the vital point value.

4. Testing: The final stage includes testing the obtained values to check whether the system delivers the correct results. This is obtained from re-training the model if needed until a best-performing model is derived. In case of any miscalculations or when the retrieved result is not satisfactory, the process starts right from the preliminary stage of data collection.

Tools and software used:

Computer vision

Deep learning

CNN

Transfer learning

Features

Determining the exact height of a candidate in the process of PET has a lot of complications where prominent ones are inaccuracy, corruption, or no proof of data that has been collected. With this, the process of height calculation has become a tricky exercise.