Contact person: Video Traffic Counter Software

Phone:Show

Email:Show

Location: Tingalpa, Australia

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"I need a Windows-based desktop application that ingests roadside video and returns reliable traffic counts every 15 minutes. The program must detect Cars, Trucks, Motorcycles—and the many additional classes listed in my attached guide—then log each detected vehicle against the movement direction it takes.

The workflow I picture is straightforward: I open a video file, draw or edit virtual lines/arrows on a preview pane to mark the directions of travel, hit Start, and let the software do the rest. While the video is processed it should output each 15-minute block completed, to an automatically growing spreadsheet (CSV or XLSX) that shows the timestamp, direction, and individual counts per vehicle class. Spreadsheet pages may be arranged according to vehicle type, for example.

For any unidentified, or poorly identified vehicles (i.e. identification varies across frames), the application should output the relevant video frames for review by an operator, who can then use the UI of the app to identify the vehicle from a list and indicate the direction taken.

A clean graphical interface for creating and saving those direction templates is essential, since each camera view is different. Please keep direction entry entirely interactive—no hard-coded compass points.

Deliverables
• Compiled Windows installer (or portable EXE)
• Source code with brief build instructions
• Example spreadsheet produced from my sample footage
• Short user guide explaining how to set up new camera views and add future vehicle classes from the guide

Acceptance criteria
• ≥90 % detection accuracy on the sample videos I provide
• Correct 15-minute aggregation boundaries regardless of clip start time
• Spreadsheet matches on-screen totals exactly

If you prefer OpenCV, TensorFlow, or another library, that’s fine as long as the end user sees a single, self-contained application. Let me know any clarifications you need and the libraries you expect to use.

While using YOLO, and perhaps ChatGPT, might be your first thought - be aware that the former is unlikely to provide the accuracy required without extensive training and/or a different approach might be needed to get a reasonably wide range of vehicle types. See the attached guide for a list of the many vehicle types the project looks to identify - or, at least, as many types as possible.

You need to mention in your offer how many types you think can be identified by your approach in the deliverables, and what accuracy you expect for passenger cars and pedestrians." (client-provided description)