Heterogeneous Traffic Count Dataset, Pune (Jan 2023)
This dataset provides directional, vehicle-class-wise traffic counts derived from real-time CCTV footage recorded at three signalized intersections in Pune, Maharashtra, India — Alankar Chowk, Jehangir Chowk, and RTO Chowk — over a continuous seven-day period from 11 to 18 January 2023. Raw video was processed through a YOLO-based detection and multi-object tracking pipeline (with non-maximum suppression) to identify, classify, and count vehicles by type — car, motorbike, bus, and truck — moving in four directions (UP, DOWN, LEFT, RIGHT) at each junction approach camera. Recording sessions varied by camera and day rather than following a fixed schedule; vehicle counts are logged at approximately 5-minute intervals, with per-camera coverage detailed in the accompanying README. Each junction is monitored by two to three fixed cameras, with CSV files named by camera ID and recording date (e.g., a2_11.csv = Camera 2, Alankar Chowk, 11 January 2023). A full data dictionary and camera/naming-convention reference are included in the accompanying README and Data Descriptor files. The dataset contains only aggregated numerical vehicle counts — no video, images, or personally identifiable information is included. It is intended to support research in urban traffic flow analysis, congestion modelling, signal-timing optimization, vehicle-flow forecasting, SUMO-based traffic simulation, and computer-vision benchmarking for traffic surveillance systems. Data source: Traffic Police Commissioner, Pune. Processing infrastructure: GPU-based high-performance computing cluster (4x Tesla V100), Department of Computer Engineering, COEP Technological University, Pune, procured under TEQIP-III (World Bank project). Licensed under CC BY 4.0.