The study of roadside visual algorithm in vehicle-road cooperative based on YOLOX and KPP-DeepSORT

Visual inspection was an important technology for the roadside perception of vehicle-road cooperative. For the difficulty of keeping balance between detection precision and computation efficiency by common visual perception algorithms, a new visual processing method based on YOLOX and KPP-DeepSORT was presented. First, YOLOX was used for multi-channel image recognition, then considering the orderliness of the traffic stream, K-Means++ was introduced to preprocess DeepSORT for reducing tracking delay. Data showed that the proposed method had high accuracy for detecting and tracking the pedestrians and vehicles. Especially in rush-hour crossroad, it performed more efficiently than some other common algorithms, so that had extensive application prospect in Internet of Vehicles.