AI Vehicle Type and Traffic Flow Recognition: Moving Annual Traffic Volume Surveys from Manual Counting to Automation

Every year, the regional maintenance branches of Taiwan's Highway Bureau (Ministry of Transportation and Communications) conduct traffic volume surveys on provincial highways, and local governments have their own traffic flow survey needs. Traffic impact assessments for building development projects likewise require traffic volume data for intersections and road segments. These survey results are an important basis for road planning, signal design, and traffic improvement.

However, data such as intersection turning movements and vehicle type composition has traditionally required surveyors to count vehicles one by one, either on site or from recorded video. For long-duration, multi-location surveys, this demands a large amount of manpower, and manual counts can also be affected by differences in how individual surveyors classify vehicles.

HITech's Vehicle Type / Traffic Flow / Road Condition Recognition helps consulting firms and survey contractors that undertake traffic volume surveys analyze recorded video automatically with AI image recognition, reducing the workload of manual vehicle-by-vehicle counting so staff can focus on on-site setup, data checking, and follow-up analysis.

Seven Vehicle Types, Mappable to Highway Traffic Volume Categories

The system recognizes seven vehicle types—motorcycles, passenger cars, buses, small trucks, large trucks, semi-trailers, and full trailers—with a vehicle detection rate of over 99.5%, and can classify vehicles and count traffic according to survey requirements.

Highway traffic volume statistics commonly use categories such as small vehicles, buses, large trucks, full trailers, semi-trailers, and motorcycles. HITech's system further divides small vehicles into passenger cars and small trucks, and the two can be combined into a single "small vehicle" count when a survey requires it.

Note that AI image recognition classifies vehicles mainly by their visual appearance in the video; it does not determine vehicle type from registration records, gross vehicle weight, or motor vehicle office data.

Traffic Flow, Turning Movements, and Average Speed: Multiple Parameters from One Recording

Beyond vehicle type and traffic counts, the system can track vehicle trajectories and directions of travel for intersection turning movement analysis, and calculate the average speed in a given area as a reference for assessing congestion and traffic operations.

A single road recording can be analyzed for multiple survey parameters, reducing the need to schedule separate manual counts for each statistic.

72 Hours of Video, Recognized Frame by Frame in About 24 Hours

Traffic volume surveys often require several consecutive days of recording, and across multiple survey locations the volume of video adds up quickly.

HITech's system performs batch AI video analysis at over 90 FPS. Taking 30 FPS video as an example, 72 hours of footage can be recognized frame by frame in about 24 hours, greatly shortening the time needed to process large volumes of recordings.

Actual processing performance varies with the hardware used, video resolution, frame rate, and the analysis items selected.

Extensive Road Video Data and Field Validation Across Many Scenes

HITech has built up long-term image recognition experience across different roads and traffic scenes. Its data covers more than 1,500 scenes, 108,000 hours of video, and 75 million vehicles, and is continually used for model development, testing, and validation of recognition performance.

Taiwan's roads have a high share of motorcycles and mixed car-and-motorcycle traffic. Situations such as large vehicles blocking motorcycles from view, or several motorcycles passing closely together or overlapping, can make image recognition more difficult.

Actual recognition results are therefore affected by camera placement, shooting angle, resolution, lighting, occlusion, and traffic density. For formal traffic volume surveys, we recommend assessing the filming conditions at each survey location and planning the recording in advance, so the footage is well suited to AI analysis.

AI Image Recognition in Traffic Data Collection

AI image recognition is becoming one of the technical tools for collecting traffic data. For annual traffic volume surveys that involve long, multi-location recordings, automated video analysis can reduce the need for large-scale manual counting and establish a more consistent data processing workflow.

HITech offers AI video analysis for vehicle type recognition, traffic counting, intersection turning movements, and road conditions. We can assess survey locations, recording conditions, and the statistics required, and propose a suitable AI vehicle type and traffic flow recognition solution.

Consulting firms and survey contractors working on traffic volume surveys, traffic impact assessments, and transportation planning are welcome to contact us. For feature details and real recognition videos, visit the Vehicle Type / Traffic Flow / Road Condition Recognition product page.