Object detection from aerial view in Drone Programming - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding challenges in aerial object detection

Which of the following is the main challenge when detecting objects from aerial drone images?

A Objects appear very small and densely packed, making them hard to distinguish

B Objects are always occluded by trees and buildings, so detection is impossible

C Objects move too fast for drones to capture clear images

D Aerial images have low resolution compared to satellite images

💡 Hint

Think about how objects look from high above and how that affects detection.

From aerial views, objects often appear very small and close together, which makes it difficult for models to separate and identify them accurately.

❓ Predict Output

intermediate

Output of bounding box format conversion

What is the output of this Python code that converts bounding boxes from (x_center, y_center, width, height) to (x_min, y_min, x_max, y_max)?

Drone Programming

def convert_bbox(cx, cy, w, h):
    x_min = cx - w / 2
    y_min = cy - h / 2
    x_max = cx + w / 2
    y_max = cy + h / 2
    return (x_min, y_min, x_max, y_max)

print(convert_bbox(50, 50, 20, 10))

A (30.0, 45.0, 70.0, 55.0)

B (50, 50, 70, 60)

C (60.0, 55.0, 40.0, 45.0)

D (40.0, 45.0, 60.0, 55.0)

💡 Hint

Remember the center minus half width is the minimum x coordinate.

The function calculates the corners by subtracting and adding half the width and height from the center coordinates.

❓ Model Choice

advanced

Best model architecture for small object detection in aerial images

Which model architecture is best suited for detecting small objects in aerial drone images?

A Recurrent Neural Network (RNN) designed for text sequences

B Simple CNN with only fully connected layers

C YOLOv5 with multi-scale feature fusion and attention modules

D Linear regression model predicting object size

💡 Hint

Look for a model designed for object detection with features to handle small objects.

YOLOv5 with multi-scale features and attention helps detect small objects by focusing on different image scales and important regions.

❓ Hyperparameter

advanced

Choosing the right anchor box sizes for aerial object detection

When training an object detection model on aerial images with mostly small vehicles, which anchor box sizes should you choose?

A Very large anchor boxes to cover the entire image

B Small anchor boxes matching the average vehicle size in pixels

C Anchor boxes with zero width and height

D Random anchor boxes unrelated to object sizes

💡 Hint

Anchor boxes should match the typical object sizes to help the model learn better.

Choosing anchor boxes close to the actual object sizes improves detection accuracy by providing better initial guesses.

❓ Metrics

expert

Interpreting mAP in aerial object detection evaluation

If a model trained for aerial object detection achieves a mean Average Precision (mAP) of 0.75 at IoU threshold 0.5, what does this mean?

A On average, the model correctly detects 75% of objects with bounding boxes overlapping ground truth by at least 50%

B The model detects 75% of objects but with no regard to bounding box accuracy

C The model's predictions overlap ground truth by exactly 75% for all objects

D The model has 75% classification accuracy on object types

💡 Hint

mAP combines precision and recall considering bounding box overlap.

mAP at IoU 0.5 means the model's bounding boxes overlap ground truth by at least 50%, and on average, it detects 75% of objects correctly.