An Unified Framework for Motorbike Counting and Detecting in Traffic Videos

Motorbike is an interesting subject for both the object counting and object detecting problem. In developing countries, motorbike is the most popular vehicle. In traffic, knowledge of motorbikes count will reveal the overall density while knowing of motorbikes positions will reveal the direction of vehicles flows. For each of those tasks, we usually need a different approach. But, it will be wasteful if these approaches were to run separately. Therefore, we propose a combination of counting and detection which can reveal both the motorbikes count and position. To achieve this, our model reuses a result from the counting task, the density map, in the detection task. Using this density map will help us reduce the area for detection. The experiments show that our proposed method is faster than when the counting task and detection task were to run separately. It is also faster than the detection process running alone.