TrafGoAVI: the Demo Interface for TrafGo VTS SDK
Main Interface
| The following image shows the main interface of TrafGoAVI 1.0.1.4. Observe that the interface of TrafGoAVI consists of many regions to demonstrate the rich functionalities and the flexibilities that TrafGo VTA SDK provides to the developers. |
 |
| The following shows the interface to load and unload lane structures. A YangSky lane structure is a region mask consisting of a set of polygons that can be edited using the LaneEdt.exe program and save as a *.lan type file. When clicking on the “Load Lane” button, a *.lan file loading interface will appear and the user can load different types of lanes into the program. The user can also unload any types of lane structure from the program by clicking the “Unload Lane” button. |
 |
This is a lane loading *.lan file interface:  |
This is the drop list of different lane type that can be chosen from:  |
The lane division of a simulated road with a LN_SHOULER lane and a LN_MAINLANE.  |
1. Automatic Environment Learning and Adapting.When a new AVI video clip is loaded, the automatic environment learning functionality will be initiated and the following shows the learning progress. This function can be triggers whena. The initialization of the entire system. b. The camera shift to a new pose and angle c. Dramatic change of illumination conditions and/or weather conditions. This function can be controlled manually by using control signals or can be triggered from some built-in smart environment assessment modules. |
 |
2. Automatic Lane Configuration Learning.TrafGo has a build-in automatic lane structure learning function to automatically learning the lane configuration of a road. Based on different applications, TrafGo can learn the main lane, the side walk and the reverse lane. It is also provide the ability to learn the sub-lanes within the main road if the learning period is long enough. |
This picture shows the early stage of lane learning. At this stage, TrafGo tries to figure out how many lanes in this traffic road and the thick green lines indicates the first few guess of the lanes. As time goes by, TrafGo will find the most significant lanes.
 |
This picture shows the lane learning is in progress. Observe that only four most significant lanes are ready for the next stage of learning. The weaker candidates in the earlier learning stages were removed from the learning space.  |
To make it easy to observe that behaviors of \TrafGoAVI, the user can press the “Pause” button to pause the video clip, 
and to continue, the reader can press the “Play” button  .
|
This picture shows the final results of the lane learning. Observe that only the three separated traffic lanes are marked in this case; namely, the side walk, the main lanes and the reverse lanes.  |
3. Pedestrian Detection.TrafGo provides a set of cognitive model based pedestrian detection and tracking functions as demonstrated by the following image. Observe that in the results window we used a solid white block to indicate the present of an pedestrian. |
 |
This picture shows the dimension of attributes that TrafGo used to distinguish a pedestrian from vehicles. All these attributes are implemented by using computational nouns and computational verbs. The reader can manipulate these parameters and see the changes of the detecting behaviors.
 |
4. Breakdown Detection.If a vehicle was breakdown on any portion of the road, TrafGo used a solid read block to mark such a event as show in this simulation: |
 |
A pedestrian marked by a red solid block because he is stay at the sidewalk to fix his bike.  |
5. Reverse Driving.To detect the reverse driving, we first need to define the legal driving directions first. As shown in the following image, TrafGo use a simple GUI to implement this. |
 |
In this case, the legal driving direction are defined as from top to bottom:  |
This is the simple GUI to choosing different legal driving directions.  |
When a reverse driving is detected, a thick bright blue rectangle will be drawn at the moving object.  |
6. General Parameters. |
 |
|
BkThresh: the threshold to distinguish the background and the foreground. Big value will result in separation of a single big vehicle. Small value will result in merge of very closed vehicle. LearnRate: the rate to learn the environment. BkLearnT: the length of the learning period of the environment. For complex environments you might need a longer learning period. SzThresh: the threshold for the area in pixels of objects that we need to detect. A small value can results in high sensitivity to small objects while sensitive to noise. BdObjTh: reserved. MvStaticR: the sensitivity to distinguish a moving objects to a static/breakdown objects. A big value results in longer distinguish time period. RePeriod: the time period in frames for distinguish a static object from a slowly moving objects. A small value results in sensitive detection to static object while might view slowly moved objects as static. |
 |
|
Height: the height in pixels for an object to be classified as a pedestrian. This should be an average value or most likely values of the pedestrian height.
Speed: the typical moving speed of a pedestrian. H/W ratio: the ratio of Height and Width of a pedestrian figure. Degree: the minimum truth value of an object to be classified as a pedestrian. |