Traffic Camera Vehicle Detection and Counting Dashboard

Cities and transportation departments make road investment and signal timing decisions with surprisingly little data, because manual traffic counts are expensive and in-road sensors are costly to install and maintain. Existing traffic cameras already watch most busy intersections, making software the cheapest path to reliable vehicle counts. The intern builds a Streamlit dashboard that analyzes traffic footage using two generations of computer vision: a classical detector combining HOG features with an SVM classifier, and a modern YOLO deep learning model running on PyTorch. The application processes video with OpenCV, draws detections onto the footage, counts vehicles crossing defined zones, and aggregates results with NumPy into traffic analytics such as per-interval volumes and detection confidence summaries. A comparison view puts the classical and neural pipelines side by side so users can evaluate accuracy against processing cost on the same clip. Through the build, the intern learns how detection models are evaluated in a practical setting rather than on benchmark leaderboards, develops skills in video processing and counting logic such as line-crossing checks, and demonstrates the ability to deliver a decision-support dashboard for a real civic use case.