We proposes an affinity matching multi-target vehicle tracking system based on minimal cost linear cost allocation. The vehicle tracking unit tracking system’s goal is to capture scene records from cameras mounted on a moving self-vehicle. Compared to other low-speed tracking applications (such as traditional pedestrian tracking), vehicle tracking on road scenes and images acquired from the camera of a moving self vehicle magnify the problem of larger bounding box geometry changes. This vehicle tracking unit disturbance occurs in many tracking scenarios, such as when a high-speed object approaches from an opposite lane. Because autonomous driving algorithms need to use processing resources in an efficient manner, our research is particularly focused on developing computational light weights even when it meets the requirements for computing complex tasks such as positioning, object detection, occupancy grid update, sensor fusion, and trajectory planning. The vehicle tracking unit On-line multi-object tracking models and benchmarks.
Multi-target tracking and its ability to predict surrounding dynamic traffic scenarios play a vital role in autonomous driving. The performance enhancement of convolutional neural networks tracking units for vehicles in terms of run time and detection accuracy has created “detection and tracking” paradigms. Networks that provide higher accuracy and a lower number of false negatives need to embed more complexity, need to adjust a large number of parameters and more processing requirements.
In view of the positioning, automatic detection, sensor fusion, occupancy grid update, trajectory planning, dynamic modeling, and control tasks used in autonomous driving , , a wide range of applications require highly efficient and accurate solutions. Whether using Lidar point clouds, stereo objects, single-camera image sensors, or online or batch processing methods, multi-target trackers are divided. We propose a lightweight online multi-target vehicle tracking method vehicle gps tracking units, which is an online “detection and tracking” multi-target vehicle tracker that relies on a single camera.
Our method is superior to the most advanced method. At the same time, our method is based on the “Car” level of the KITTI object tracking benchmark in the first three measurement results of the MOTA, MOTP, and MT tests. Run-time performance (0.03 seconds) will be tested and cross-checked. Compared with other methods,tracking unit for car the speed is increased by 6 to 20 times, which realizes the calculation of “lightweight” multi-vehicle tracking for autonomous driving.
The vehicle tracking unit achieved runtime performance enables trajectory planning based on the movement of surrounding vehicles in a challenging tracking scenario. Run-time performance along acceptable levels of tracking metrics enables the computing resources of autonomous vehicles to be used for other time-critical tasks.