System Requirements

Mandatory Performance Requirements
The system will:

  • M.P.1 Capture video stream at a speed of at least 30 FPS
    • The Data Capture Unit will capture video at a minimum rate of 30 frames per second.
  • M.P.2 Capture 100% view of intersection
    • Given an intersection with n separate roads, Data Capture Unit must cover all the roads. To achieve this, there can be one or more sensors pointing in different directions.
  • M.P.3 Have detection recall of more than 75%
    • Out of all the actors present in the scene, the system will detect at least 75% of actors (cars and pedestrians) present in the captured data.
  • M.P.4 Have detection precision of more than 75%
    • Out of all the detected actors, the system will correctly classify at least 75% of cars and pedestrians.
  • M.P.5 MOTA (multi-object tracking accuracy) should be atleast 40\% and MOTP(multi-object tracking precision) should be atleast 40\%.
    • Given the association threshold of 2 metres, the tracker should be able to associate at least 40% of the cars present in the scene correctly.
    • The tracker should ensure that the average deviation of the cars’ predictions is not more than 40% against the ground truth.
  • M.P.6 Have a simulation speed of minimum 10 FPS supporting at least 3 simultaneous agents
    • The system will simulate a traffic scenario in the simulator at a speed of at least 10 frames per second when at least 3 behavioral agents are spawned.
  • M.P.7 Observed Parameters (list) matching configured parameters with 20% tolerance

Mandatory Non-Functional Requirements
The system shall:

  • M.N.1 Have a modular design for components
    • The subsystems should be modular enough such that changing a subsystem or a part of it does not require modification in other subsystems.
  • M.N.2 Have a compact and portable sensor suite.
    • The system requires a wide variety of data. The Data Capture Unit needs to be mounted on different intersections and roads to collect that. So the Data Capture Unit needs to be compact and portable.
  • M.N.3 Have well documented code and APIs.
    • Team shall maintain a detailed document of the APIs being implemented during the course of the project, so that it is easy for the user to develop the system further.
  • M.N.4 Have well documented code and APIs.
    • Allow for tuning aggression parameters of models.
        The behavioral model developed should be able to model certain desirable behavioral nuances parametrically.

Desirable Performance Requirements
The system will:

  • D.P.1 Have detection recall of 90% or more
    • The system will detect 90% or more of the cars and pedestrians present in the captured data.
  • D.P.2 Have detection precision of 90% or more
    • The system will correctly classify 90% or more of cars and pedestrians.
  • D.P.3 Have a simulation speed of 30 FPS or more
    • The system will simulate a traffic scenario in the simulator at 30 frames per second or more.
  • D.P.4 Support 5 or more simultaneous actors at 15 fps.
    • In order to simulate more complex behaviors, the system will simultaneously simulate 5 or more agents in a traffic scenario.
  • D.P. 5 The average L2 distance between the executed trajectory and the ground truth should be less than 3.5m

Desirable Non-Functional Requirements
The system shall:

  • D.N.1 Be scalable to different simulation environments
    • The system should not be limited to one self driving simulator. Hence the system shall be scalable to different simulation environments.
  • D.N.2 Be less expensive to train
    • The learning complexity increases manifolds with number of actors, hence computational cost and time for training should scale reasonably with the number of actors.
  • D.N.3 Comply to IP64 standards for Sensor module
    • The Data Capture Unit has to be mounted in different places with different weather conditions. Hence it is desirable to make it IP64 complaint.