CV Models (Segmentation, Detection, Tracking)

Related API: ccai9012.yolo_utils · ccai9012.svi_utils

Overview

Category: Perception & Prediction from Visual Data

Modular Components: - Object Detection/Tracking with YOLO - Semantic Segmentation Model - Trajectory Extraction - Visualization

Use Cases

• What factors influence walking behavior?
• How does visual cleanliness (graffiti, trash, lighting) relate to perceived safety?
• Can we predict CO2 emission using the SVIs?

Code Examples

Pedestrian Behavior Analysis in Public Spaces

Content: - Detect pedestrians using YOLO - Track movement using DeepSORT - Analyze flow, dwell time, and walkability

Dataset: - Webcam data - Source: https://www.skylinewebcams.com/en.html

Required Packages: YOLOv5, OpenCV, DeepSORT, numpy, matplotlib

Identify pedestrian location and generate footprint heatmap with tracking.

SVI-Based Housing Price Prediction

Content: - Use subjective perception scores (e.g., cleanliness, greenery) on SVI - Combine CV scoring with regression to predict housing price - Visual quality → real estate value linkage

Datasets: - Google Street View Imagery (SVI) from Google Map API - California housing price dataset from sklearn.datasets

Required Packages: OpenCV, scikit-learn, pandas, matplotlib, PyTorch

SVI-based housing price estimation. Nouriani, A., Lemke, L., 2022. Vision-based housing price estimation using interior, exterior & satellite images. Intelligent Systems with Applications 14, 200081. https://doi.org/10.1016/j.iswa.2022.200081.