Artificial intelligence and computer vision powered driving-performance assessment

An artificial intelligence (AI) based computer vision (CV) method for detecting pose and position estimation for objects vehicles (204), moving or parked in a driving yard and/or parking lot (200), the method comprising:

• one or more routines stored in a memory, operable to detect one or more vehicles (204) from one or more markers installed on the one or more vehicles in a series of images or videos using Artificial Neural Network;
• capturing, by one or more or network of cameras (202), one or more visuals of one or more markers installed on a movable object that is configured to move in a predefined area;
• detecting, by a processor in communication with the one or more or network of cameras (202), a pose of each of the one or more markers from the captured one or more visuals;
• determining, by the processor, one or more parameters related to a movement of the movable object in the predefined area based on detecting the pose of the one or more markers;
• where such parameters are 2D location, speed and direction of the vehicle (204) and
• implementing the fusion of marker locations together from each camera within the network of cameras (202) to estimate the final vehicle location;
• implement a calibration technique to match frames from the cameras to the a 3-dimensional (3D) world;
• plot regions of interest on the 3D world onto a 2-dimensional (2D) layouts of the 3D world;
• and
• automatically detecting, by the processor, one or more driving violations committed by the movable object based on the determined one or more parameters;
• implemented by one or more processors coupled to the memory and operable to execute the one or more routines to:
• wherein the one or more processors are further configured to:
• synchronize frames from the cameras (202) such that all frames from the one or more cameras (202) are of the same time stamp to generate the most accurate pose and position of the objects or vehicles;
• synchronize frames from the cameras (202) using multi-threading, wherein a synchronizer thread and a processing thread are initiated on the launch of the program and separate capture threads are launched that start capturing the frames at their respective speed;
• enable the synchronizer thread to estimate a slowest camera (202) based on a first frame received from the slowest camera (202), wherein the first frame forms an index key;
• and send the index key to the processing thread, which starts querying frames for the remaining cameras (202).