Research On The Vehicle Detection Technology In Fire Engine Access

The main purpose of a fire engine access vehicle detection system based on Internet of Things is to effectively avoid the fire disaster leading by the fire control passageway blockage. The system combines sensors and artificial intelligence by Internet of Things, and adopts the digital image processing and pattern intelligent recognition technologies to detect the vehicles automatically. When a vehicle occupies the fire engine access, the system realizes vehicle detection and on-lite alarm operation automatically. Also, the management personnel is prompted to clear the vehicle occupancy by messages and the fire department receives the vehicle occupancy information, which is considered to do the punishment and accountability. The vehicle detection algorithm is the core technology of the fire engine access detection system. Thus, to find a suitable fire engine access vehicle detection algorithm is significant.Aiming at the low detection rate and poor robustness existing in the vehicle detection algorithm of the fire engine access vehicle detection system, this paper considers the impact on the vehicle detection caused by various interference factors, such as the weather variation, the ambient environment, etc. This paper presents a vehicle bottom shadow splitting method based on self-adaptive threshold and a vehicle detection algorithm based on multi-features integration cascading classifier. The main contributions are as follows:① In the generation phase of the interesting region of the vehicle, this paper proposes a self-adaptive threshold segmentation method to get the shadows at the bottom of vehicles. Firstly, the road area of an image is extracted, excluding the impact of non-lane area objects on the simulation results. Secondly, the gradation value of the extracted road area is sampled to calculate the segmentation threshold of bottom shadows. Then the lower border of the shadow is extracted by the edge fetching method based on the pixels’ varying rate. Finally, the interested region of vehicle is constructed according to the low border of the shadow.② In the extraction and fusion stage of vehicles multi-feature, this paper first introduces the core features, edge features and textural features. Then, from the prospective of character fusion, this paper analyzes comparatively the integration effect of core features, edge features and textural features, resulted from the multi-features integration method based on between-class variance, Mahalanobis-distance and Fisher criterion. The simulation results for the fusion of the components in the vehicle sequence images show that the effect of the multi-features fusion method based on the Fisher criterion is the best.③ In the validation phase of the interesting region of the vehicle, a vehicle detection method based on multi-features fusion cascading classifier is proposed. First, the corner features, edge features and textural features are integrated by a multi-feature integration algorithm based on the Fisher criterion, and the strong classifier is generated utilizing the Adaboost algorithm by training the characteristics derived by the Fisher criterion and the Harr-like characteristics derived by the “integral image”. Then, the interesting region of the vehicle is verified by the trained cascading classifier.The experimental results show that compared with the traditional methods of vehicle detection, the proposed method can self-adaptively detect vehicles in the fire engine access with a better detection rate and false positive rate in different climatic conditions.