Existing robotic grasping operations are usually performed under well-illuminated conditions with clear object details and high regional contrast. At the same time， for low-light conditions caused by night and occlusion， where the objects’ visual features are weak， the detection accuracies of existing robotic grasp detection models decrease dramatically. In order to improve the representation ability of sparse and weak grasp features in low-light scenarios， a grasp detection model incorporating visual feature enhancement mechanism was proposed to use the visual enhancement sub-task to impose feature enhancement constraints on grasp detection. In grasp detection module， the U-Net like encoder-decoder structure was adopted to achieve efficient feature fusion. In low-light enhancement module， the texture and color information was respectively extracted from local and global level， thereby balancing the object details and visual effect in feature enhancement. In addition， two low-light grasp datasets called low-light Cornell dataset and low-light Jacquard dataset were constructed as new benchmark dataset of low-light grasp and used to conduct the comparative experiments. Experimental results show that the accuracies of the proposed low-light grasp detection model are 95.5% and 87.4% on the benchmark datasets respectively， which are 11.1， 1.2 percentage points higher on low-light Cornell dataset and 5.5， 5.0 percentage points higher on low-light Jacquard dataset than those of the existing grasp detection models， including Generative Grasping Convolutional Neural Network （GG-CNN）， and Generative Residual Convolutional Neural Network （GR-ConvNet）， indicating that the proposed model has good grasp detection performance.

Aiming at the problem of large feature reconstruction error and local constraint loss between features in fast low rank coding algorithm, an enhanced local constraint fast low rank coding algorithm was put forward. Firstly, the clustering algorithm was used to cluster the features in the image, and obtain the local similarity feature set and the corresponding clustering center. Secondly, the K visual words were found by using the K Nearest Neighbor (KNN) strategy in the visual dictionary, and then the K visual words were combined into the corresponding visual dictionary. Finally, the corresponding feature code of the local similarity feature set was obtained by using the fast low rank coding algorithm. On Scene-15 and Caltech-101 image datasets, the classification accuracy of the modified algorithm was improved by 4% to 8% compared with the original fast low rank coding algorithm, and the coding efficiency was improved by 5 to 6 times compared with sparse coding. The experimental results demonstrate that the modified algorithm can make local similarity features have similar codes, so as to express the image content more accurately, and improve the classification accuracy and coding efficiency.

Aiming at such interferences as false detection and missed detection which can't be overcome by the existing virtual loop detection algorithm in multilane traffic flow detection, a novel traffic flow detection algorithm based on adaptive virtual loop was put forward. According to the image binarization principle, quadratic estimation was adopted in the foreground detection part of the Visual Background extractor (ViBe) algorithm, and the background updating mechanism was changed. A new improved ViBe algorithm was presented to achieve the purposes of rapidly eliminating the ghost and completing the foreground object extraction. Then, the fixed detection area was set on the road, and the mobile virtual loop was established or canceled according to the moving target trajectory of fixed detection area. The traffic flow algorithm based on virtual loop was further used to achieve traffic flow statistics. Three different scenarios:no vehicle lane change with 4 lanes, vehicle lane change with 2 lanes, vehicle lane change with 3 lanes and sudden environmental change, were chosen for experiments and the traffic flow detection accuracy of the proposed algorithm was 8.9, 25 and 16.6 percentage points higher than that of the traditional virtual loop detection algorithm. The experimental results show that the proposed algorithm is more suitable for multilane traffic flow detection.