The 4th AI City Challenge

Abstract

The 6th edition of the AI City Challenge specifically focuses on problems in two domains where there is tremendous unlocked potential at the intersection of computer vision and artificial intelligence: Intelligent Traffic Systems (ITS), and brick and mortar retail businesses. The four challenge tracks of the 2022 AI City Challenge received participation requests from 254 teams across 27 countries. Track 1 addressed city-scale multi-target multicamera (MTMC) vehicle tracking. Track 2 addressed natural-language-based vehicle track retrieval. Track 3 was a brand new track for naturalistic driving analysis, where the data were captured by several cameras mounted inside the vehicle focusing on driver safety, and the task was to classify driver actions. Track 4 was another new track aiming to achieve retail store automated checkout using only a single view camera. We released two leader boards for submissions based on different methods, including a public leader board for the contest, where no use of external data is allowed, and a general leader board for all submitted results. The top performance of participating teams established strong baselines and even outperformed the state-ofthe-art in the proposed challenge tracks.

Prediction of Pedestrian Crossing Intentions at Intersections Based on Long Short-Term Memory Recurrent Neural Network

“…Ten correspondences of points in the traffic scene were extracted from images and Google Maps using the OpenCV package ( 39 ). Each of the point correspondence could form two rows of matrix A. Singular value decomposition (SVD) could be employed to obtain solution of this formula ( 40 – 42 ). After getting matrix h , the image coordinates could be converted to the world coordinates through the inverse matrix of h .…”

Vehicle-Rear: A New Dataset to Explore Feature Fusion for Vehicle Identification Using Convolutional Neural Networks

“…In this sense, we remark that the above datasets -as well as others available in the literature -have a different purpose (a) VeRi-776 [61] (b) VERI-Wild [8] (c) VehicleID [10] (d) CityFlowV2 [9] FIGURE 13. Vehicle rear images of four public datasets: in the VeRi-776 dataset (a), most license plates are not legible and the authors did not provide any annotations for the plates; in the VERI-Wild, VehicleID and CityFlowV2 datasets (b-d), the license plates were redacted due to privacy restrictions.…”

“…Content may change prior to final publication. the literature, namely, VeRi-776 [61], VERI-Wild [8], Ve-hicleID [10], and CityFlowV2 [9]. An overview of these datasets is presented in Table 8.…”

“…We emphasize that even though in [6], [39] the authors claimed that they extended the VeRi-776 dataset with license plate annotations, these annotations were not made available due to privacy restrictions (according to the first author of [6], [39], [61]). In the VERI-Wild [8], CityFlowV2 [9] and VehicleID [10] datasets, on the other hand, it is not even possible to exploit information from the license plate regions for vehicle identification, as they were purposely redacted in all images (with a black bounding box) by the respective authors because of privacy restrictions. For the record, CityFlowV2 is an updated version -with refined annotations -of CityFlow [63].…”

Estimating Vehicle and Pedestrian Activity from Town and City Traffic Cameras

“…With each challenge, increasing focus has been given to analyzing multi-camera networks that simulate existing large-scale urban systems. For example, in 2020, participants were encouraged to compete in one or more of the following four categories: Multi-Class Multi-Movement Vehicle Counting, City-Scale Multi-Camera Vehicle Re-Identification, City-Scale Multi-Camera Vehicle Tracking, and Traffic Anomaly Detection [ 40 ]. Among successful experiments, object detection models from the YOLO series and Faster-RCNN were utilized for their superior speeds and performance on consumer-grade GPUs.…”