This project reads an image and based on the dimensions of a reference object; it finds the dimensions of other objects in a scene. The reference object must be the leftmost, topmost object in the scene. In this case, a box of dimension 2cm x 2cm is taken as a reference object. For any other reference object, we can provide the actual width of the object.

Prerequisites:

1. Python 3
2. OpenCV
3. NumPy
4. SciPy
5. VS Code, Google Colab (Any IDE)

Constraints

1. Shadows can cause incorrect prediction. This can be avoided by making sure there is enough light.
2. To get accurate object boundary, dark background is used.

Algorithm

1. Image pre-processing
• Read an image and convert it to grayscale
• Blur the image using Gaussian Kernel to remove unnecessary edges/ reduce noises.
• Edge detection using Canny edge detector is performed on the blurred image using a threshold value.
• Create a kernel for morphological operations (dilation and erosion) using NumPy.
• Dilate the edges obtained from Canny edge detection to make them more contiguous.
• Erode the dilated image to further refine the edges.

2. Reference object
• Calculate how many pixels are there per metric (centimeter is used here)

3. Compute results
• Draw bounding boxes around each object and calculate its height and width