Optical flow concept in Computer Vision - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What does optical flow represent in video analysis?

In simple terms, what does optical flow describe when analyzing a video sequence?

A The color changes of objects over time

B The pattern of apparent motion of objects between consecutive frames

C The brightness level of each pixel in a single frame

D The audio signal associated with the video

💡 Hint

Think about how objects move from one frame to the next in a video.

Optical flow captures how pixels move between frames, showing motion patterns.

❓ Predict Output

intermediate

Output of Optical Flow Calculation Code

What is the shape of the output array from the following optical flow calculation using OpenCV's calcOpticalFlowFarneback function?

Computer Vision

import cv2
import numpy as np
prev_frame = np.zeros((100, 100), dtype=np.uint8)
next_frame = np.zeros((100, 100), dtype=np.uint8)
flow = cv2.calcOpticalFlowFarneback(prev_frame, next_frame, None, 0.5, 3, 15, 3, 5, 1.2, 0)
output_shape = flow.shape

A (100, 100, 2)

B (100, 100)

C (2, 100, 100)

D (100, 2)

💡 Hint

Optical flow returns a vector for each pixel indicating motion in x and y directions.

The output is a 3D array with height and width matching the input frames, and 2 channels for horizontal and vertical flow components.

❓ Model Choice

advanced

Choosing a Model for Dense Optical Flow Estimation

Which model architecture is best suited for estimating dense optical flow between two images?

A A convolutional neural network designed to output a 2-channel flow field for each pixel

B A simple feedforward network with a single output neuron

C A recurrent neural network designed for text generation

D A clustering algorithm like K-means

💡 Hint

Dense optical flow requires per-pixel motion vectors.

CNNs can learn spatial features and output a flow vector per pixel, making them ideal for dense optical flow.

❓ Hyperparameter

advanced

Effect of Pyramid Scale Parameter in Optical Flow Calculation

In the Farneback optical flow algorithm, what is the effect of increasing the pyramid scale parameter (pyr_scale)?

A It decreases the number of pyramid levels, reducing accuracy

B It controls the threshold for motion vector magnitude

C It changes the color space of the input images

D It increases the resolution of the image pyramid, allowing finer motion detection but increasing computation time

💡 Hint

Think about how image pyramids help detect motion at different scales.

A larger pyr_scale means the pyramid images are closer in size, increasing resolution and detail but requiring more computation.

❓ Metrics

expert

Evaluating Optical Flow Accuracy with Endpoint Error

Given ground truth flow vectors and predicted flow vectors, how is the average endpoint error (EPE) calculated?

A By measuring the correlation coefficient between predicted and true flow magnitudes

B By counting the number of pixels where predicted flow equals ground truth exactly

C By computing the average Euclidean distance between predicted and true flow vectors over all pixels

D By calculating the mean squared error of pixel intensities

💡 Hint

Endpoint error measures how far off the predicted motion vectors are from the true ones.

EPE is the average length of the vector difference between predicted and ground truth flow vectors, showing average prediction error.