The tracking algorithm plays a crucial role in various application scenarios, especially in ADAS product, where it is used to track the same object and predict its movement direction.

In applications of surveillance systems and autonomous driving, the tracking algorithm is indispensable and facing several challenges,

Occlusion issue: When the target object is obscured by other objects, tracking becomes difficult. This requires the algorithm to predict target object's location to effectively handle occlusion.

Environmental changes: External factors like lighting and weather conditions will affect tracking performance.

Target deformation: The tracked object may change its shape or pose.

Multi-object tracking: The algorithm must be capable of simultaneously tracking multiple objects and categories.

In order to overcome these challenges, we proposed a tracking algorithm with the following advantages:

Object Tracking: In the field of autonomous driving, this algorithm can automatically identify and track road users, such as vehicles or pedestrians, thereby enhancing driving safety.

Tracking Stability: It effectively addresses the issue of bounding box flickering, achieving stable object tracking.

Multi-object Tracking: It is capable of performing specialized tracking tasks for different categories, such as pedestrians, cars, or motorcycles.

Result